A fan bearing fault diagnosis method, device, equipment and readable storage medium

Abstract

This invention relates to the field of wind turbine operation and maintenance technology, and discloses a method, device, equipment, and readable storage medium for diagnosing wind turbine bearing faults. The method provided by this invention preprocesses the raw vibration response data acquired in real time, mapping it to a unified spatiotemporal grid to obtain a standardized spatiotemporal data matrix. Based on this spatiotemporal data matrix, and combined with a trained inverse physical information neural network model, a quasi-Newton iteration method with kinetic momentum is used to perform fault parameter inversion optimization, accurately solving for the optimal fault parameters that conform to the bearing dynamics, without relying on massive labeled samples. Finally, the optimal fault parameters are analyzed using a physical mapping evaluation method, directly associating the fault parameters with the bearing's geometric kinematics and fault dynamics, thereby achieving quantitative identification of the depth, width, and spatial location of wind turbine bearing defects and qualitative identification of fault types. This achieves the goal of qualitative identification and quantitative assessment of wind turbine bearing faults under small sample conditions.

Description

Technical Field

[0001] This invention relates to the field of wind turbine operation and maintenance technology, specifically to a method, apparatus, equipment, and readable storage medium for diagnosing wind turbine bearing faults. Background Technology

[0002] Wind turbines are key equipment in the field of clean energy, and their operational reliability is crucial to the stable supply of energy. Core rotating components such as main shaft bearings and generator bearings are prone to fatigue, wear, and spalling failures due to long-term exposure to complex alternating loads. Once they fail, they will cause the entire machine to shut down, resulting in significant economic losses. Therefore, accurate early fault diagnosis of wind turbine bearings is of great engineering significance.

[0003] The disclosed wind turbine bearing fault diagnosis methods in related technologies include physical model-based methods and data-driven methods. Physical model-based methods are based on the dynamic differential equations of the bearing and simulate fault response by establishing a vibration propagation model. However, the actual operating conditions of wind turbine bearings are complex, with time-varying speeds and loads, and structural parameters such as internal stiffness and damping are difficult to obtain accurately, easily leading to a mismatch between the model and the actual system. Furthermore, the computational cost of model solving is high. While data-driven methods can directly learn fault characteristics from monitoring data without complex physical modeling, this method heavily relies on a large number of labeled fault samples, and the prediction results may violate basic dynamic laws. Therefore, the wind turbine bearing fault diagnosis methods disclosed in related technologies are difficult to achieve qualitative identification and quantitative assessment of wind turbine bearing faults under small sample conditions. Summary of the Invention

[0004] This invention provides a method, apparatus, device, and readable storage medium for diagnosing wind turbine bearing faults, in order to solve the problem that the wind turbine bearing fault diagnosis methods disclosed in related technologies are difficult to achieve qualitative identification and quantitative assessment of wind turbine bearing faults under small sample conditions.

[0005] In a first aspect, the present invention provides a method for diagnosing wind turbine bearing faults, the method comprising: Based on the raw data of the vibration response of the wind turbine bearing acquired in real time, a preprocessing method is used to map it to a unified spatiotemporal grid to obtain a preprocessed spatiotemporal data matrix. The preprocessing method includes spatiotemporal discretization, multi-source data synchronization, noise reduction filtering, normalization and bilinear interpolation. Based on the preprocessed spatiotemporal data matrix, combined with the trained inverse physical information neural network model, the fault parameters are inverted and optimized using the quasi-Newton iteration method with kinetic momentum to obtain the optimal fault parameters of the wind turbine bearing. Based on the optimal fault parameters, the physical mapping evaluation method is used for analysis to obtain the fault diagnosis results of the wind turbine bearing; the fault diagnosis results of the wind turbine bearing include: quantitative identification results of the depth, width, and spatial location of the wind turbine bearing defects, and qualitative identification results of the fault type.

[0006] Through the above implementation method, the real-time acquired vibration response raw data is first preprocessed and mapped to a unified spatiotemporal grid to eliminate the spatiotemporal heterogeneity, noise interference, and sensor measurement differences of the original vibration response data, resulting in a standardized spatiotemporal data matrix. Then, based on the spatiotemporal data matrix, combined with a trained inverse physical information neural network model, the fault parameter inversion optimization solution is carried out using the quasi-Newton iteration method with kinetic momentum. The optimal fault parameters that conform to the bearing dynamics law are accurately solved without relying on massive labeled samples. Finally, based on the obtained optimal fault parameters, the physical mapping evaluation method is used for analysis, directly associating the fault parameters with the bearing geometric kinematics and fault dynamics law, thereby realizing the quantitative identification of the depth, width, and spatial location of wind turbine bearing defects and the qualitative identification of fault types, achieving the goal of qualitative identification and quantitative evaluation of wind turbine bearing faults under small sample conditions.

[0007] In one optional implementation, based on the preprocessed spatiotemporal data matrix and combined with the trained inverse physical information neural network model, the fault parameter inversion optimization solution is obtained using the quasi-Newton iterative method with kinetic momentum to obtain the optimal fault parameters of the wind turbine bearing, including: To optimize the network parameters of the trained inverse physical information neural network model, all weight matrices and bias vectors are fixed, and the set of wind turbine bearing fault features is taken as the variable to be optimized. Based on the preprocessed spatiotemporal data matrix, it is input into the inverse physical information neural network model, and the multi-fidelity total loss function is minimized as the optimization objective to construct the fault parameter inversion optimization function. Based on the fault parameter inversion optimization function, the quasi-Newton iteration method with driving force is used to iterate the fault feature set of the wind turbine bearing until the iteration termination condition is met; the iteration termination condition includes the number of iterations reaching the preset total number of iterations or the difference between the total loss function values ​​of two adjacent iterations being less than the preset convergence threshold. Based on the iterative set of wind turbine bearing fault characteristics, the optimal fault parameters of the wind turbine bearing are determined.

[0008] Through the above implementation method, the optimal network parameters of the trained inverse physical information neural network model are fixed, and only the set of wind turbine bearing fault features is used as the variable to be optimized. This avoids redundant calculations in retraining the network parameters, ensuring the efficiency of parameter inversion. Simultaneously, relying on the physical constraints and feature mapping capabilities of the trained model, a precise model foundation is laid for solving the fault parameters. Then, the preprocessed spatiotemporal data matrix is ​​input into the model, and an inversion optimization function is constructed with minimizing the multi-fidelity total loss function as the optimization objective. This ensures that the optimal fault parameter solution process simultaneously conforms to measured vibration data, the physical laws of bearing dynamics, and boundary constraints, thus optimizing the process. At the objective level, the physical consistency and data fit of the solution results are guaranteed. Subsequently, the quasi-Newton iteration method with driving force is used to iteratively solve the fault feature set. Combined with the dual termination conditions of the number of iterations reaching a preset value and the difference in total loss between adjacent iterations being less than the convergence threshold, it avoids iterative overfitting, ensures the convergence and accuracy of the solution results, accelerates convergence, reduces iterative oscillations, and improves the overall efficiency of parameter inversion. Finally, the optimal fault parameters are obtained from the iterated fault feature set, realizing accurate back-inference from vibration response data to core fault parameters, and obtaining the optimal fault parameters that can truly reflect the actual fault characteristics of the wind turbine bearing.

[0009] In one optional implementation, the multi-fidelity total loss function includes: Based on the predicted vibration response output by the inverse physical information neural network model and the actual observation data of the spatiotemporal data matrix, the data residual term, physical residual term and boundary residual term are obtained by using the residual calculation method. Based on the data residuals, physical residuals, and boundary residuals, the neural tangent kernel algorithm is used to analyze the gradient norm change trend of the corresponding residuals during backpropagation, and the weight coefficients of the corresponding residuals in the total loss function are adjusted to obtain the multi-fidelity total loss function.

[0010] Through the above implementation method, the residuals of the vibration response predicted by the inverse physical information neural network model and the actual observation data are calculated first, and the data residuals, physical residuals, and boundary residuals are obtained respectively. This allows the total loss function to simultaneously take into account the fit of the inverse physical information neural network model to the measured data, the conformity with the physical laws of bearing dynamics, and the satisfaction of the bearing housing boundary conditions, thus achieving a deep integration of physical mechanisms and data-driven approaches. Then, relying on the neural tangent kernel matrix trace operation method, the gradient norm change trend of each residual term in backpropagation is analyzed and the weight coefficients are dynamically adjusted. This enables the total loss function to adaptively balance the importance of data fitting, physical constraints, and boundary matching according to the actual process of model training and parameter inversion. This effectively solves the optimization imbalance problem caused by the differences in the dimensions and gradient scale of different residual terms, ensuring that the optimization direction of the final multi-fidelity total loss function is more in line with the actual needs of fault diagnosis.

[0011] In one optional implementation, the multi-fidelity total loss function includes: , , , , in, Represents physical residuals; Indicates data residuals; Indicates boundary residuals; The weighting coefficients represent the physical residuals; The weighting coefficients represent the data residuals; The weighting coefficients representing the boundary residuals; This represents the number of random sampling points used to calculate the physical residual; This represents the bearing seat displacement response predicted by the inverse physical information neural network; Indicates the density of the bearing housing material; This represents the second-order partial derivative of the bearing seat displacement response predicted by the inverse physics information neural network model with respect to time t. Represents the stress divergence term; For fault impact source terms, used to represent the fault impact source term in the first... Spatial location related to each fault and time The magnitude of the impact force at the point; Indicates the number of observed data points; This represents the vibration response predicted by the inverse physics information neural network model at the observation location. and time The value at; This represents the actual observed vibration response; Indicates the number of boundary condition points; This represents the vibration response predicted by the inverse physics information neural network model at the boundary location. and time The value at; This indicates the vibration response specified by the boundary conditions.

[0012] In one optional implementation, the raw data of the real-time acquired wind turbine bearing vibration response is preprocessed and mapped to a unified spatiotemporal grid to obtain a preprocessed spatiotemporal data matrix, including: Based on the real-time acquired raw data of wind turbine bearing vibration response, the cross-correlation function between each channel signal and the reference channel signal is calculated using the cross-correlation method, and the time delay corresponding to the peak value is extracted to obtain multi-source vibration data after time synchronization. Based on the time-synchronized multi-source vibration data, standardized vibration data is obtained by using a combination of variational mode decomposition denoising, bandpass filtering feature extraction, and amplitude normalization. Based on the standardized vibration data and combined with the preset spatiotemporal grid parameters, the discrete measurement point data is mapped to each node of the spatiotemporal grid using bilinear interpolation to obtain the preprocessed spatiotemporal data matrix.

[0013] Through the above implementation method, the time synchronization of multi-channel vibration data is achieved by using the cross-correlation method, which accurately eliminates the time delay deviation in the multi-sensor acquisition process and ensures the consistency of data from different measuring points in the time dimension. Then, through the combined processing of variational mode decomposition denoising, bandpass filtering feature extraction and amplitude normalization, noise interference in the original data is effectively filtered out, fault characteristic related frequency band signals are retained, and the measurement amplitude differences between different sensors are eliminated, resulting in standardized vibration data that only reflects the fault characteristics of the wind turbine bearing. Finally, combined with preset spatiotemporal grid parameters, the standardized data of discrete measuring points are mapped to the spatiotemporal grid nodes of the entire field using the bilinear interpolation method, completing the full-field vibration data of the radial plane of the bearing housing, forming a spatiotemporal data matrix that can accurately and comprehensively reflect the full-field vibration characteristics of the wind turbine bearing.

[0014] In one optional implementation, the step of analyzing the optimal fault parameters using a physical mapping evaluation method to obtain the fault diagnosis result of the wind turbine bearing includes: Based on the optimal fault parameters, quantitative feature parameters are extracted from the optimal fault parameters to obtain quantitative identification results of the defect geometric dimensions and fault spatial location of the wind turbine bearing, and qualitative identification results of the fault type. Based on the quantitative identification results of the defect geometry and fault spatial location, and the qualitative identification results of the fault type, combined with the preset fault severity grading standard, the overall severity of the fault is comprehensively evaluated, and the fault diagnosis results and fault severity grading results of the wind turbine bearing are output.

[0015] Through the above implementation method, quantitative feature parameters are first accurately extracted from the optimal fault parameters, and then directly mapped to obtain quantitative identification results of the geometric dimensions of wind turbine bearing defects and the spatial location of faults, as well as qualitative identification results of fault types. This realizes the physical analysis of fault characteristics from parameters to actual fault representation, allowing the diagnostic results to have both quantitative accuracy and clear physical meaning, without black box nature and verifiability. Then, by integrating the quantitative and qualitative diagnostic results and combining them with the preset fault severity grading standards, a comprehensive assessment of the overall severity of the fault is carried out. This not only achieves accurate determination of wind turbine bearing faults, but also completes the hierarchical classification of the degree of fault development, providing wind power equipment operation and maintenance personnel with comprehensive, intuitive, and practical fault handling basis.

[0016] In one optional implementation, the optimal fault parameters include: defect depth, defect opening width, fault impact amplitude, defect center coordinates, and optimal characteristic frequency; the step of extracting quantitative characteristic parameters from the optimal fault parameters to obtain quantitative identification results of the defect geometry and fault spatial location of the wind turbine bearing, and qualitative identification results of the fault type, based on the optimal fault parameters, includes: Based on the defect depth and defect opening width, the geometric dimensions of the wind turbine bearing defect are used as quantitative identification results, and the fault impact amplitude is used as an auxiliary quantitative indicator of the fault impact intensity. Based on the coordinates of the defect center, combined with the coordinate system of the radial plane of the wind turbine bearing housing, the spatial position of the defect on the bearing component is analyzed, the specific component and circumferential orientation of the fault are determined, and a quantitative identification result of the fault spatial position is obtained. Based on the optimal characteristic frequency, combined with the bearing geometric kinematics formula, and substituting the inherent structural parameters and real-time operating parameters of the wind turbine bearing, the matching degree between the optimal characteristic frequency and the theoretical value of the characteristic frequency for different fault types of the bearing is verified, and the qualitative identification result of the fault type is determined.

[0017] Through the above implementation methods, the defect depth and defect opening width are directly used as quantitative identification results of defect geometry, and the fault impact amplitude is used as an auxiliary quantitative indicator of fault impact intensity, thus achieving an intuitive quantitative characterization of fault severity. Based on the defect center coordinates combined with the bearing housing radial plane coordinate system, the specific component and circumferential orientation of the fault are accurately located, obtaining a quantitative identification result of the fault spatial location. Then, based on the optimal characteristic frequency, the bearing geometric kinematic formula is matched and verified with the bearing's inherent structural parameters and real-time operating parameters to determine the qualitative identification result of the fault type. The entire analysis process is completed entirely based on physical laws, and the output diagnostic results have both quantitative accuracy and physical interpretability, providing an accurate, reliable, and comprehensive basis for wind turbine bearing operation and maintenance decisions.

[0018] Secondly, the present invention provides a wind turbine bearing fault diagnosis device, the device comprising: The data preprocessing module is used to map the raw data of the vibration response of the wind turbine bearing acquired in real time to a unified spatiotemporal grid using preprocessing methods to obtain a preprocessed spatiotemporal data matrix. The preprocessing methods include spatiotemporal discretization, multi-source data synchronization, noise reduction filtering, normalization, and bilinear interpolation. The fault inversion module is used to perform fault parameter inversion optimization based on the preprocessed spatiotemporal data matrix and the trained inverse physical information neural network model, using the quasi-Newton iteration method with momentum, to obtain the optimal fault parameters of the wind turbine bearing. The result output module is used to analyze the optimal fault parameters using a physical mapping evaluation method to obtain the fault diagnosis results of the wind turbine bearing. The fault diagnosis results of the wind turbine bearing include: quantitative identification results of the depth, width, and spatial location of the wind turbine bearing defects, and qualitative identification results of the fault type.

[0019] Thirdly, the present invention provides an electronic device, comprising: a memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions to perform the wind turbine bearing fault diagnosis method of the first aspect or any corresponding embodiment described above.

[0020] Fourthly, the present invention provides a computer-readable storage medium storing computer instructions for causing a computer to execute the wind turbine bearing fault diagnosis method of the first aspect or any corresponding embodiment described above. Attached Figure Description

[0021] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0022] Figure 1 This is a schematic flowchart of the first method for diagnosing wind turbine bearing faults according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the second process of the wind turbine bearing fault diagnosis method according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the third process of the wind turbine bearing fault diagnosis method according to an embodiment of the present invention; Figure 4 This is a schematic diagram of the fourth process of the wind turbine bearing fault diagnosis method according to an embodiment of the present invention; Figure 5 This is a structural block diagram of a wind turbine bearing fault diagnosis device according to an embodiment of the present invention; Figure 6 This is a schematic diagram of the hardware structure of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] It is understood that before using the technical solutions disclosed in the various embodiments of the present invention, users should be informed of the types, scope of use, and usage scenarios of the personal information involved in the present invention and their authorization should be obtained in accordance with relevant laws and regulations through appropriate means.

[0025] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0026] The wind turbine bearing fault diagnosis methods disclosed in related technologies include: 1. The physical model-based approach uses the dynamic differential equations of bearings as its theoretical foundation and establishes a vibration propagation model to simulate fault response. However, the actual working environment of wind turbine bearings is complex, with time-varying speeds and loads, and internal structural parameters (such as stiffness and damping) are difficult to obtain accurately. This leads to a significant deviation between the established physical model and the actual system, i.e., a "model mismatch" problem. Furthermore, the model solution process is complex and computationally costly, making it difficult to meet the real-time requirements of online diagnostics.

[0027] 2. Data-driven methods employ deep learning techniques (such as convolutional neural networks and recurrent neural networks) to directly learn fault characteristics and patterns from massive amounts of monitoring data (such as vibration signals). This method avoids complex physical modeling and exhibits strong feature extraction capabilities with large amounts of labeled data. However, it still has the following drawbacks: In real-world production processes, it is difficult to obtain a large amount of labeled data (i.e., "fault samples") covering all types and severity of faults, which is extremely costly in engineering practice. The learning process of the model lacks physical constraints, and its prediction results may violate basic physical laws (such as energy conservation and dynamic equations), resulting in poor generalization ability under working conditions not covered by training data, or even absurd diagnostic results. Data-driven methods can only achieve qualitative identification of fault types (such as inner ring faults and outer ring faults), but cannot accurately quantify the severity of faults (such as crack length and spalling area).

[0028] In summary, the wind turbine bearing fault diagnosis methods disclosed in related technologies are difficult to achieve qualitative identification and quantitative assessment of wind turbine bearing faults under small sample conditions. To overcome these shortcomings, this invention provides a wind turbine bearing fault diagnosis method. It preprocesses the real-time acquired vibration response raw data and maps it to a unified spatiotemporal grid, eliminating the spatiotemporal heterogeneity, noise interference, and sensor measurement differences of the raw vibration response data, resulting in a standardized spatiotemporal data matrix. Based on this spatiotemporal data matrix, and combined with a trained inverse physical information neural network model, a quasi-Newton iterative method with kinetic momentum is used to perform fault parameter inversion optimization, accurately solving for the optimal fault parameters that conform to the bearing dynamics, without relying on massive labeled samples. Finally, based on the obtained optimal fault parameters, a physical mapping evaluation method is used for analysis, directly associating the fault parameters with the bearing's geometric kinematics and fault dynamics, thereby achieving quantitative identification of the depth, width, and spatial location of wind turbine bearing defects and qualitative identification of fault types, achieving the goal of qualitative identification and quantitative assessment of wind turbine bearing faults under small sample conditions.

[0029] According to an embodiment of the present invention, a method for diagnosing wind turbine bearing faults is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0030] This embodiment provides a method for diagnosing wind turbine bearing faults, which can be used in the operation and maintenance server terminal of a wind power station. Figure 1 This is a flowchart of a wind turbine bearing fault diagnosis method according to an embodiment of the present invention, such as... Figure 1 As shown, the process includes the following steps: S101, based on the real-time acquired raw data of the wind turbine bearing vibration response, a preprocessing method is used to map it to a unified spatiotemporal grid to obtain a preprocessed spatiotemporal data matrix; the preprocessing method includes spatiotemporal discretization, multi-source data synchronization, noise reduction filtering, normalization and bilinear interpolation.

[0031] The raw vibration response data of a wind turbine bearing is the unprocessed vibration signal data collected in real time by monitoring equipment such as accelerometers during the operation of the wind turbine bearing due to its own rotation, component friction, or faults. In this embodiment of the invention, the raw vibration response data of the wind turbine bearing is collected by three unidirectional accelerometers evenly arranged along the circumferential direction on the outer surface of the wind turbine bearing housing, with a 120° interval between adjacent unidirectional accelerometers, used to capture the transient impact signal generated when the rolling elements of the wind turbine bearing pass through local defects.

[0032] Preprocessing methods are data processing techniques used to transform the messy raw data of wind turbine bearing vibration response into standardized data suitable for neural network models. These methods include spatiotemporal discretization, multi-source data synchronization, noise reduction filtering, normalization, and bilinear interpolation, which are used to eliminate data interference in the raw data of wind turbine bearing vibration response, unify the data format, and complete the whole field data.

[0033] Spatiotemporal discretization involves splitting continuous time-dimensional vibration signals into discrete spatiotemporal node data by dividing continuous spatial-dimensional bearing housing regions into discrete spatiotemporal node data, which facilitates subsequent calculations and mesh mapping.

[0034] Specifically, the outer surface of the bearing housing is divided into multiple small areas, and the center position of each area is denoted as . Preset spatial step size Then set the sensor sampling frequency to a uniform value. Continuously record the transient impact signals generated by the rolling elements of the wind turbine bearing passing through defects. ,in For sensor coordinates, For the corresponding moment, the time step .

[0035] Multi-source data synchronization can be implemented by using cross-correlation method for time synchronization. This method eliminates time delay deviations caused by factors such as equipment and installation, and ensures the consistency of vibration data collected from different measuring points at the same time in the time dimension.

[0036] Multi-source data synchronization methods specifically include: From signals collected by multiple sensors, one signal is selected as a reference. For example, the signal from the first accelerometer sensor is selected as the reference signal, denoted as [reference signal]. ; For signals from other sensors , Its relationship with the reference signal cross-correlation function Defined as: , in, It's a time delay. It refers to the data collection time; Then use the cross-correlation function The peak position determines the time delay between the signals from the two sensors. Specifically, find the way to The largest Value, that is: , Finally, the calculated time delay is used. Signals from other sensors Perform a time shift to align it with the reference signal. Alignment, expressed as follows: .

[0037] Denoising filtering is a processing method used to eliminate irrelevant noise in the original vibration data and extract fault feature signals. By using variational mode decomposition, bandpass filters and other methods, it filters out invalid noise such as environmental interference and equipment clutter, and retains characteristic frequency band signals related to wind turbine bearing faults, thereby improving the effectiveness of the data.

[0038] Normalization, on the other hand, is a processing method that eliminates the differences in measurement amplitude between different sensors and maps vibration data to a fixed numerical range. This ensures that multi-source vibration data collected by different sensors are consistent in terms of dimensions and amplitude, avoids data deviations caused by differences in sensor hardware, and improves data consistency.

[0039] Bilinear interpolation is a numerical interpolation method that maps vibration data from discrete sensor measurement points to all nodes of a preset full-field spatiotemporal grid for the bearing housing. It calculates the vibration values ​​of unknown nodes in the grid using known measurement point data, thus completing the vibration data of the entire bearing housing and realizing the transformation from "single-point data" to "full-field data".

[0040] The unified spatiotemporal grid is a discrete spatiotemporal coordinate grid pre-defined for the radial plane of the wind turbine bearing housing, divided according to a fixed spatial step size (e.g., 0.5 mm) and a time step size (e.g., 0.05 ms) to adapt to the input of the inverse physical information neural network model. All vibration data are ultimately mapped to the nodes of this grid, ensuring the standardization of data format and dimensions.

[0041] The spatiotemporal data matrix is ​​a standardized matrix formed by preprocessing the original vibration data. It uses spatiotemporal grid nodes as indices and vibration response values ​​as elements. Rows or columns correspond to nodes in space or time. Each element in the matrix represents the vibration response value of a bearing housing at a certain spatial location and time.

[0042] The preprocessed vibration signal is mapped onto a unified spatiotemporal grid. The specific implementation steps are as follows: a1. The acquired signal is denoised and its features are extracted. First, variational mode decomposition is used to decompose the original signal into multiple modal components, and then the modal components containing the main fault characteristics are selected. Next, filters (such as bandpass filters) are used to further remove noise and other irrelevant frequency components, retaining the intermediate frequency components related to the fault characteristics. Finally, the signal is normalized to the same amplitude range to reduce measurement differences between different sensors.

[0043] a2, based on the set spatial step size and time step Construct a spatiotemporal grid. Spatial grid points are... The time grid points are ,in =1,2,…, , =1,2,…, , , These are the number of spatial grid points and the number of temporal grid points, respectively.

[0044] a3 uses bilinear interpolation to interpolate the preprocessed signal onto the grid nodes.

[0045] Assuming four neighboring grid points are known Bilinear interpolation can be performed using the following formula: , in, These are interpolation coefficients, calculated based on the positional relationship between the target point and known points, and are expressed as follows: , , , .

[0046] a4, constructing a spatiotemporal data matrix from the interpolated data. Each element Indicates spatial location and time point Vibration response value at the location.

[0047] S102, based on the preprocessed spatiotemporal data matrix and combined with the trained inverse physical information neural network model, the fault parameters are inverted and optimized using the quasi-Newton iteration method with driving momentum to obtain the optimal fault parameters of the wind turbine bearing.

[0048] The inverse physics information neural network model is a specialized model that integrates the physical mechanism of bearing dynamics with neural networks. Unlike traditional neural networks that only fit data, it embeds the physical laws such as the linear elastic dynamics control equations and boundary conditions of the bearing into the network structure and loss function. Its core function is to solve the "inverse problem" of deriving fault parameters from vibration response data, rather than the "direct problem" of predicting vibration response from fault parameters.

[0049] To facilitate the conversion of the defect geometric parameters of wind turbine bearings into source terms of the control equations for inputting into the inverse physics information neural network model, the construction of the inverse physics neural network model includes: The radial plane of the bearing housing is taken as the research domain. Under the plane strain assumption, the displacement field is selected. For the basic unknowns, where Represents the Cartesian coordinates within the domain (unit: m). Time (unit: seconds).

[0050] Based on linear elasticity theory, the governing equations are as follows: , ,

[0051] in, The material density of the bearing housing ( ); Represents the stress tensor; It is a fourth-order isotropic elastic tensor, derived from Young's modulus. Compared with Poisson Decide, All can be 1, 2, or 3, corresponding to three spatial directions; Represents the linear strain tensor; This indicates the fault impact source term, representing the transient impact generated when the rolling element passes through a local defect.

[0052] The fault impact source term is specifically represented as follows: , in, This represents the set of fault features to be identified. , Indicates the impact amplitude, in units of , Indicates the coordinates of the defect center. This indicates the width of the defect opening, in meters (m). This indicates the depth of the defect, in meters (m). The characteristic frequency is represented by the frequency coupled with the bearing's geometric kinematics. , Indicates the number of rolling elements. The rotor rotation frequency, in Hz. The diameter of the rolling element is in meters (m). The contact angle is expressed in rad. It is a unit step function used to spatially truncate the impact zone.

[0053] The aforementioned control equations facilitate the association between the defect geometric parameters of the wind turbine bearing and the source terms of the control equations, providing a physically consistent low-dimensional input space for the subsequent neural network.

[0054] In order to achieve the extraction of fault parameters Vibration response The efficient mapping proposed in this invention utilizes the inverse physical information neural network model based on fault parameters. (Including fault impact amplitude) Defect center location Defect opening width ,depth and characteristic frequencies Using ) as input, it generates at the spatiotemporal grid nodes Full-field displacement response It includes: an initial feature mapping layer, a dual-branch feature extraction layer, a gated attention feature fusion layer, a physical perception correction layer, and a multi-task learning output layer.

[0055] The initial feature mapping layer is used to map fault parameters. The initial feature field is mapped through two fully connected layers, with tanh as the activation function, and the output dimensions are 128 and 256, respectively, to obtain high-dimensional feature vectors. .

[0056] , , in, , It is a weight matrix. , It is the bias vector.

[0057] The dual-branch feature extraction layer employs a dual-branch structure to analyze features in both the temporal and spatial dimensions. Multi-scale one-dimensional convolutional layers extract the dynamic characteristics of the time series, capturing features at different time scales. Convolutional kernel sizes are 3, 5, and 7, with a stride of 1 and 64 output channels. Dimensionality reduction is achieved through a max-pooling layer to obtain the temporal feature vector. The expression is as follows: , The spatial distribution characteristics of the sensor are then analyzed by combining two-dimensional convolution and graph neural networks (GNNs). Two-dimensional convolution extracts local spatial features; the first convolution kernel is 3×3 with a stride of 1 and 32 output channels; the second convolution kernel is 5×5 with a stride of 2 and 64 output channels. The graph neural network models global correlations based on the sensor's topology to obtain spatial feature vectors. The expression is as follows: , The gated attention feature fusion layer is used to extract feature vectors from the temporal and spatial branches. and The fused feature vector is obtained by dynamically fusing features through a gating attention mechanism. The gated unit uses the sigmoid activation function to dynamically adjust the weights of temporal and spatial features, generating a fused feature vector, as shown below: , , in, Represents the gate vector; This represents the sigmoid activation function; Represents the weight matrix; Represents the bias vector; This indicates element-wise multiplication.

[0058] The physical perception correction layer dynamically adjusts the network prediction results based on the physical residual calculation formula. The specific implementation is as follows: (1) Based on the governing equations, calculate the residual between the vibration response predicted by the network and the physical laws. That is, calculate the physical residual. : , in, The stress divergence term is represented by the stress divergence term, which is due to the elastic tensor. and the stress distribution caused by the vibration response gradient, The initial predicted oscillating response of the main branch of the multi-task learning output layer originates from the deconvolution layer's fusion of feature vectors. The mapping, i.e. ; For fault impact source terms, it indicates the first fault impact term. Spatial location related to each fault and time The magnitude of the impact force at the point.

[0059] (2) Construct a small neural network using a multilayer perceptron (MLP) to achieve physical residuals. Using the input, we obtain the predicted correction term. It satisfies the following: , (3) Generate the corrected predicted vibration response It satisfies the following: .

[0060] The multi-task learning output layer employs a multi-task learning mechanism: the main branch uses a deconvolution layer to fuse the feature vectors. Mapping onto a spatiotemporal grid generates the initial vibration response. The auxiliary branch directly predicts fault parameters through the fully connected layer. ,satisfy: .

[0061] The momentum-based quasi-Newton iteration method is an efficient numerical optimization solution method. It reduces computation and improves convergence speed by approximating the Hessian matrix (second derivative matrix) instead of direct differentiation. It also uses the momentum term to introduce the update direction of the previous iteration, reducing iteration oscillations and accelerating convergence. In this embodiment, it is used to iteratively optimize the fault feature set and solve for the optimal fault parameters.

[0062] The fault parameter inversion optimization solution is a process of deriving unknown wind turbine bearing fault characteristic parameters from a known spatiotemporal data matrix through numerical optimization methods. It relies on inverse physical information neural networks and quasi-Newton iteration methods for driving forces to achieve accurate inversion from "data characteristics" to "fault parameters".

[0063] The optimal fault parameters are estimated values ​​of the fault feature set that best fit the actual fault state of the wind turbine bearing, obtained through fault parameter inversion optimization. They are expressed as follows: It is used to accurately reflect the core information of bearing failure, such as impact amplitude, defect location, geometric dimensions, and characteristic frequency.

[0064] S103, based on the optimal fault parameters, the physical mapping evaluation method is used for analysis to obtain the fault diagnosis results of the wind turbine bearing; the fault diagnosis results of the wind turbine bearing include: quantitative identification results of the depth, width, and spatial location of the wind turbine bearing defects, and qualitative identification results of the fault type.

[0065] The physical mapping evaluation method is an analytical method that transforms the obtained optimal fault parameters into intuitive fault diagnosis results based on the objective physical laws of bearing geometry and fault dynamics. Its core is to realize the physical meaning mapping from "fault parameters" to "actual fault representation".

[0066] The fault diagnosis result is the final output of the wind turbine bearing fault status judgment result, which combines quantitative and qualitative identification results. The quantitative identification result is the geometric feature result of the wind turbine bearing fault, which is directly mapped through the optimal fault parameters and characterized by specific numerical values, including defect depth, width, and spatial location. The qualitative identification result is the qualitative judgment result of the wind turbine bearing fault type, which is obtained by combining the optimal fault parameters (such as characteristic frequency and defect location) with the analysis of the physical laws of the bearing, and clarifies the component of the bearing where the fault occurs (such as inner ring, outer ring, or rolling element) and the fault mode (such as crack, spalling, or fatigue).

[0067] This invention provides a method for diagnosing wind turbine bearing faults. It preprocesses real-time acquired vibration response data and maps it to a unified spatiotemporal grid, eliminating spatiotemporal heterogeneity, noise interference, and sensor measurement differences, resulting in a standardized spatiotemporal data matrix. Based on this matrix, and combined with a trained inverse physical information neural network model, a quasi-Newton iterative method using kinetic momentum is employed to optimize and solve for fault parameters. This accurately determines the optimal fault parameters that conform to bearing dynamics, without relying on massive labeled samples. Finally, based on the obtained optimal fault parameters, a physical mapping evaluation method is used for analysis, directly linking the fault parameters to the bearing's geometric kinematics and fault dynamics. This enables quantitative identification of the depth, width, and spatial location of wind turbine bearing defects, as well as qualitative identification of fault types, achieving the goal of qualitative identification and quantitative assessment of wind turbine bearing faults under small sample conditions.

[0068] This embodiment provides a method for diagnosing wind turbine bearing faults, which can be used in the operation and maintenance server terminal of a wind power station. Figure 2 This is a flowchart of a wind turbine bearing fault diagnosis method according to an embodiment of the present invention, such as... Figure 2 As shown, the process includes the following steps: S201, based on the real-time acquired raw data of the wind turbine bearing vibration response, a preprocessing method is used to map it to a unified spatiotemporal grid to obtain a preprocessed spatiotemporal data matrix; the preprocessing method includes spatiotemporal discretization, multi-source data synchronization, noise reduction filtering, normalization and bilinear interpolation.

[0069] Specifically, S201 above includes: S2011, based on the real-time acquired raw data of the vibration response of the wind turbine bearing, uses the cross-correlation method to calculate the cross-correlation function between each channel signal and the reference channel signal and extracts the time delay corresponding to the peak value to obtain multi-source vibration data after time synchronization; S2012, Based on the multi-source vibration data after time synchronization, standardized vibration data is obtained by using a combination of variational mode decomposition denoising, bandpass filtering feature extraction and amplitude normalization. S2013, Based on the standardized vibration data and combined with the preset spatiotemporal grid parameters, the discrete measurement point data is mapped to each node of the spatiotemporal grid using bilinear interpolation to obtain the preprocessed spatiotemporal data matrix.

[0070] The cross-correlation method is used to synchronize the time of multi-channel vibration data, accurately eliminating the time delay deviation in the multi-sensor acquisition process and ensuring the consistency of data from different measuring points in the time dimension. Then, through a combination of variational mode decomposition denoising, bandpass filtering feature extraction and amplitude normalization, noise interference in the original data is effectively filtered out, fault characteristic related frequency band signals are retained, and the measurement amplitude differences between different sensors are eliminated, resulting in standardized vibration data that only reflects the fault characteristics of the wind turbine bearing. Finally, combined with preset spatiotemporal grid parameters, the standardized data of discrete measuring points are mapped to the spatiotemporal grid nodes of the entire field using bilinear interpolation, completing the full-field vibration data of the radial plane of the bearing housing, forming a spatiotemporal data matrix that can accurately and comprehensively reflect the full-field vibration characteristics of the wind turbine bearing.

[0071] S202, based on the preprocessed spatiotemporal data matrix and combined with the trained inverse physics information neural network model, the fault parameters are inverted and optimized using the quasi-Newton iterative method with driving forces to obtain the optimal fault parameters for the wind turbine bearing. For details, please refer to [link to relevant documentation]. Figure 1 S102 of the illustrated embodiment will not be described again here.

[0072] S203, based on the optimal fault parameters, the physical mapping evaluation method is used for analysis to obtain the fault diagnosis results of the wind turbine bearing; the fault diagnosis results of the wind turbine bearing include: quantitative identification results of the depth, width, and spatial location of the wind turbine bearing defects, and qualitative identification results of the fault type. For details, please refer to... Figure 1 S103 of the illustrated embodiment will not be described again here.

[0073] This invention provides a method for diagnosing wind turbine bearing faults. It preprocesses real-time acquired vibration response data and maps it to a unified spatiotemporal grid, eliminating spatiotemporal heterogeneity, noise interference, and sensor measurement differences, resulting in a standardized spatiotemporal data matrix. Based on this matrix, and combined with a trained inverse physical information neural network model, a quasi-Newton iterative method using kinetic momentum is employed to optimize and solve for fault parameters. This accurately determines the optimal fault parameters that conform to bearing dynamics, without relying on massive labeled samples. Finally, based on the obtained optimal fault parameters, a physical mapping evaluation method is used for analysis, directly linking the fault parameters to the bearing's geometric kinematics and fault dynamics. This enables quantitative identification of the depth, width, and spatial location of wind turbine bearing defects, as well as qualitative identification of fault types, achieving the goal of qualitative identification and quantitative assessment of wind turbine bearing faults under small sample conditions.

[0074] This embodiment provides a method for diagnosing wind turbine bearing faults, which can be used in the operation and maintenance server terminal of a wind power station. Figure 3 This is a flowchart of a wind turbine bearing fault diagnosis method according to an embodiment of the present invention, such as... Figure 3As shown, the process includes the following steps: S301, based on the raw data of wind turbine bearing vibration response acquired in real time, uses preprocessing methods to map it to a unified spatiotemporal grid, obtaining a preprocessed spatiotemporal data matrix; the preprocessing methods include spatiotemporal discretization, multi-source data synchronization, denoising filtering, normalization, and bilinear interpolation. For details, please refer to... Figure 1 S101 of the illustrated embodiment will not be described again here.

[0075] S302, based on the preprocessed spatiotemporal data matrix and combined with the trained inverse physics information neural network model, the fault parameters are inverted and optimized using the quasi-Newton iterative method with driving forces to obtain the optimal fault parameters for the wind turbine bearing. For details, please refer to [link to relevant documentation]. Figure 1 S102 of the illustrated embodiment will not be described again here.

[0076] Specifically, S302 above includes: S3021: For the optimal network parameters of the trained inverse physical information neural network model, fix all weight matrices and bias vectors, and take the set of wind turbine bearing fault features as variables to be optimized.

[0077] The process of training an inverse physics information neural network model includes: b1, First construct the multi-fidelity loss function: , , , , in, Represents physical residuals; Indicates data residuals; Indicates boundary residuals; The weighting coefficients represent the physical residuals; The weighting coefficients represent the data residuals; The weighting coefficients representing the boundary residuals; This represents the number of random sampling points used to calculate the physical residual; This represents the bearing seat displacement response predicted by the inverse physical information neural network; Indicates the density of the bearing housing material; This represents the second-order partial derivative of the bearing seat displacement response predicted by the inverse physics information neural network model with respect to time t. Represents the stress divergence term; For fault impact source terms, used to represent the fault impact source term in the first... Spatial location related to each fault and time The magnitude of the impact force at the point; Indicates the number of observed data points; This represents the vibration response predicted by the inverse physics information neural network model at the observation location. and time The value at; This represents the actual observed vibration response; Indicates the number of boundary condition points; This represents the vibration response predicted by the inverse physics information neural network model at the boundary location. and time The value at; This indicates the vibration response specified by the boundary conditions.

[0078] b2 employs a dynamic adjustment mechanism using the Neural Tangent Kernel (NTK) algorithm to automatically balance the weights among physical constraints, data fitting, and auxiliary tasks, including: (1) For each loss term (physical residual term, data residual term, boundary residual term), calculate the corresponding NTK submatrix. , and It satisfies the following: , , , in, Represents physical residuals For fault parameters The gradient; Represents data residuals For fault parameters The gradient; Represents boundary residuals For fault parameters The gradient; This represents the matrix transpose operation.

[0079] (2) Calculate the weighting coefficients: , , Then the value 1 is taken as the baseline. Represents the trace operation of a matrix.

[0080] (3) Apply the calculated weighting coefficients to the multifidelity loss function. This ensures that each loss term has an appropriate weight during training.

[0081] b3, Minimize the multifidelity loss function using the Adam optimization algorithm. Update network parameters, including all weight matrices. ( , , , , ), all bias vectors ( , , , , The training process includes forward propagation, loss calculation, backpropagation, and parameter update.

[0082] For example, b3 above includes: The Xavier initialization method is used to initialize all network weight matrices. and the bias vector Initialize to a zero vector; The input data is propagated forward through the network to obtain the predicted vibration response. and fault parameters The input data includes: a spatiotemporal data matrix. and known fault parameters The fault parameters are derived from the labeled dataset; The constructed loss function is used for calculation, and the gradient of the loss function with respect to the network parameters is calculated through the backpropagation algorithm. , Then, the Adam optimization algorithm is used to update the network parameters, satisfying the following: , , , Similarly, , in, This represents the first-order moment estimate, which is the exponentially weighted average of the gradient. This represents the second-order moment estimate, which is the exponentially weighted average of the squared gradients. This represents the attenuation rate estimated by the first moment, typically taken as 0.9; This represents the attenuation rate estimated by the second moment, typically taken as 0.999; It is the learning rate, used to control the step size of parameter updates, and is usually initialized to 0.001. It is a very small constant used to prevent the denominator from being zero; it is usually taken as 10. -8 ; By repeating the forward propagation, back propagation, and parameter update steps, the total loss of adjacent iterations is calculated. The relative change is less than 10 -6 Alternatively, stop training and output the optimal network parameters when the maximum number of iterations (e.g., 10,000) is reached. , .

[0083] During training, the performance of the inverse physics information neural network model is periodically evaluated on the validation set to monitor its generalization ability and prevent overfitting. If the validation loss no longer decreases significantly over five consecutive iterations, the inverse physics information neural network model is considered overfitted, and an early stopping mechanism is initiated to halt training.

[0084] S3022, Based on the preprocessed spatiotemporal data matrix, it is input into the inverse physical information neural network model, and the multi-fidelity total loss function is minimized as the optimization objective to construct the fault parameter inversion optimization function.

[0085] Specifically, the construction of the multi-fidelity total loss function includes: Based on the predicted vibration response output by the inverse physical information neural network model and the actual observation data of the spatiotemporal data matrix, the data residual term, physical residual term and boundary residual term are obtained by using the residual calculation method. Based on the data residuals, physical residuals, and boundary residuals, the neural tangent kernel algorithm is used to analyze the gradient norm change trend of the corresponding residuals during backpropagation, and the weight coefficients of the corresponding residuals in the total loss function are adjusted to obtain the multi-fidelity total loss function.

[0086] For example, the constructed multi-fidelity total loss function satisfies the following: , , , , in, Represents physical residuals; Indicates data residuals; Indicates boundary residuals; The weighting coefficients represent the physical residuals; The weighting coefficients represent the data residuals; The weighting coefficients representing the boundary residuals; This represents the number of random sampling points used to calculate the physical residual; This represents the bearing seat displacement response predicted by the inverse physical information neural network; Indicates the density of the bearing housing material; This represents the second-order partial derivative of the bearing seat displacement response predicted by the inverse physics information neural network model with respect to time t. Represents the stress divergence term; For fault impact source terms, used to represent the fault impact source term in the first... Spatial location related to each fault and time The magnitude of the impact force at the point; Indicates the number of observed data points; This represents the vibration response predicted by the inverse physics information neural network model at the observation location. and time The value at; This represents the actual observed vibration response; Indicates the number of boundary condition points; This represents the vibration response predicted by the inverse physics information neural network model at the boundary location. and time The value at; This indicates the vibration response specified by the boundary conditions.

[0087] For example, regarding the network parameters of a trained inverse physical information neural network model , Under the premise of not changing, the fault characteristic parameters Treating it as the only variable requiring optimization, the online inversion of fault parameters is achieved by solving the following optimization problem: , S3023, Based on the fault parameter inversion optimization function, the quasi-Newton iteration method with driving force is used to iterate the set of fault features of the wind turbine bearing until the iteration termination condition is met; the iteration termination condition includes the number of iterations reaching the preset total number of iterations or the difference between the total loss function values ​​of two adjacent iterations being less than the preset convergence threshold.

[0088] For example, S3023 above can be implemented to satisfy the following: , in, express The fault feature set after the next iteration; Indicates the first The fault feature set after the next iteration; This represents the iteration step size, which can take a value of 0.1. For the first The inverse of the approximate Hessian matrix in the next iteration; Represents the loss function For fault parameters The gradient; This is the momentum term coefficient, used to accelerate convergence and reduce oscillations, with a value of 0.3; express The fault feature set after the next iteration.

[0089] Meanwhile, the preset convergence threshold can be set to 10. -5 When the difference between the loss function values ​​of two consecutive iterations is less than a preset convergence threshold, the iteration is considered to have converged, and the optimization stops. After the iteration is completed, the optimal fault parameters are obtained, which facilitates the subsequent synchronous quantitative identification of the depth, width, and spatial location of wind turbine bearing defects.

[0090] S3024, Based on the iterative set of wind turbine bearing fault characteristics, determine the optimal fault parameters for the wind turbine bearing.

[0091] The quasi-Newton iteration method with driving force is used to iteratively solve the fault feature set. Combined with the dual termination conditions of the number of iterations reaching a preset value and the difference in total loss between adjacent iterations being less than the convergence threshold, it avoids iterative overfitting, ensures the convergence and accuracy of the solution results, accelerates convergence, reduces iterative oscillations, and improves the overall efficiency of parameter inversion. Finally, the optimal fault parameters are obtained from the iterated fault feature set, realizing accurate back-inference from vibration response data to core fault parameters, and obtaining the optimal fault parameters that truly reflect the actual fault characteristics of the wind turbine bearing.

[0092] S303, based on the optimal fault parameters, the physical mapping evaluation method is used for analysis to obtain the fault diagnosis results of the wind turbine bearing; the fault diagnosis results of the wind turbine bearing include: quantitative identification results of the depth, width, and spatial location of the wind turbine bearing defects, and qualitative identification results of the fault type. For details, please refer to... Figure 1 S103 of the illustrated embodiment will not be described again here.

[0093] This invention provides a method for diagnosing wind turbine bearing faults. It preprocesses real-time acquired vibration response data and maps it to a unified spatiotemporal grid, eliminating spatiotemporal heterogeneity, noise interference, and sensor measurement differences, resulting in a standardized spatiotemporal data matrix. Based on this matrix, and combined with a trained inverse physical information neural network model, a quasi-Newton iterative method using kinetic momentum is employed to optimize and solve for fault parameters. This accurately determines the optimal fault parameters that conform to bearing dynamics, without relying on massive labeled samples. Finally, based on the obtained optimal fault parameters, a physical mapping evaluation method is used for analysis, directly linking the fault parameters to the bearing's geometric kinematics and fault dynamics. This enables quantitative identification of the depth, width, and spatial location of wind turbine bearing defects, as well as qualitative identification of fault types, achieving the goal of qualitative identification and quantitative assessment of wind turbine bearing faults under small sample conditions.

[0094] This embodiment provides a method for diagnosing wind turbine bearing faults, which can be used in the operation and maintenance server terminal of a wind power station. Figure 4 This is a flowchart of a wind turbine bearing fault diagnosis method according to an embodiment of the present invention, such as... Figure 4 As shown, the process includes the following steps: S401, based on the raw data of wind turbine bearing vibration response acquired in real time, uses preprocessing methods to map it onto a unified spatiotemporal grid, obtaining a preprocessed spatiotemporal data matrix; the preprocessing methods include spatiotemporal discretization, multi-source data synchronization, denoising filtering, normalization, and bilinear interpolation. For details, please refer to [link to relevant documentation]. Figure 1 S101 of the illustrated embodiment will not be described again here.

[0095] S402, based on the preprocessed spatiotemporal data matrix and combined with the trained inverse physics information neural network model, the fault parameters are inverted and optimized using the quasi-Newton iterative method with driving forces to obtain the optimal fault parameters for the wind turbine bearing. For details, please refer to [link to relevant documentation]. Figure 1 S102 of the illustrated embodiment will not be described again here.

[0096] S403, based on the optimal fault parameters, the physical mapping evaluation method is used for analysis to obtain the fault diagnosis results of the wind turbine bearing; the fault diagnosis results of the wind turbine bearing include: quantitative identification results of the depth, width, and spatial location of the wind turbine bearing defects, and qualitative identification results of the fault type.

[0097] Specifically, S403 includes: S4031, Based on the optimal fault parameters, extract the quantitative feature parameters from the optimal fault parameters to obtain the quantitative identification results of the defect geometric dimensions and fault spatial location of the wind turbine bearing, and the qualitative identification results of the fault type, as well as the geometric and intensity quantification results. The optimal fault parameters include: defect depth, defect opening width, fault impact amplitude, defect center coordinates, and optimal characteristic frequency.

[0098] Specifically, S4031 includes: c1, based on the defect depth and defect opening width, serves as the quantitative identification result of the defect geometry of the wind turbine bearing, and the fault impact amplitude is used as an auxiliary quantitative indicator of the fault impact intensity; c2. Based on the coordinates of the defect center, combined with the coordinate system of the radial plane of the wind turbine bearing housing, the spatial position of the defect on the bearing component is analyzed, the specific component and circumferential orientation of the fault are determined, and a quantitative identification result of the fault spatial position is obtained. c3. Based on the optimal characteristic frequency, combined with the bearing geometric kinematics formula, and substituting the inherent structural parameters and real-time operating parameters of the wind turbine bearing, the matching degree between the optimal characteristic frequency and the theoretical value of the characteristic frequency for different fault types of the bearing is verified, and the qualitative identification result of the fault type is determined.

[0099] By directly using defect depth and defect opening width as quantitative identification results of defect geometry, and using fault impact amplitude as an auxiliary quantitative indicator of fault impact intensity, the severity of the fault can be intuitively and quantitatively characterized. Based on the defect center coordinates combined with the radial plane coordinate system of the bearing housing, the specific component and circumferential orientation of the fault are accurately located, resulting in a quantitative identification result of the fault spatial location. Then, based on the optimal characteristic frequency, the bearing geometric kinematic formula is matched and verified with the bearing's inherent structural parameters and real-time operating parameters to determine the qualitative identification result of the fault type. The entire analysis process is completed entirely based on physical laws, and the output diagnostic results have both quantitative accuracy and physical interpretability, providing an accurate, reliable, and comprehensive basis for wind turbine bearing operation and maintenance decisions.

[0100] S4032, combining the quantitative identification results of the defect geometry and fault spatial location with the qualitative identification results of the fault type, and in conjunction with the preset fault severity grading standard, the overall severity of the fault is comprehensively evaluated, and the fault diagnosis result and fault severity grading result of the wind turbine bearing are output.

[0101] First, quantitative feature parameters are accurately extracted from the optimal fault parameters, directly mapping to obtain quantitative identification results of the geometric dimensions of wind turbine bearing defects and the spatial location of faults, as well as qualitative identification results of fault types. This realizes the physical analysis of fault characteristics from parameters to actual fault representation, allowing the diagnostic results to have both quantitative accuracy and clear physical meaning, without black box nature and verifiability. Then, by integrating the quantitative and qualitative diagnostic results and combining them with the preset fault severity grading standards, a comprehensive assessment of the overall severity of the fault is carried out. This not only achieves accurate determination of wind turbine bearing faults, but also completes the hierarchical classification of the degree of fault development, providing wind power equipment operation and maintenance personnel with comprehensive, intuitive, and practical fault handling basis.

[0102] This invention provides a method for diagnosing wind turbine bearing faults. It preprocesses real-time acquired vibration response data and maps it to a unified spatiotemporal grid, eliminating spatiotemporal heterogeneity, noise interference, and sensor measurement differences, resulting in a standardized spatiotemporal data matrix. Based on this matrix, and combined with a trained inverse physical information neural network model, a quasi-Newton iterative method using kinetic momentum is employed to optimize and solve for fault parameters. This accurately determines the optimal fault parameters that conform to bearing dynamics, without relying on massive labeled samples. Finally, based on the obtained optimal fault parameters, a physical mapping evaluation method is used for analysis, directly linking the fault parameters to the bearing's geometric kinematics and fault dynamics. This enables quantitative identification of the depth, width, and spatial location of wind turbine bearing defects, as well as qualitative identification of fault types, achieving the goal of qualitative identification and quantitative assessment of wind turbine bearing faults under small sample conditions.

[0103] This embodiment also provides a wind turbine bearing fault diagnosis device, which is used to implement the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the term "module" can refer to a combination of software and / or hardware that performs a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0104] This embodiment provides a wind turbine bearing fault diagnosis device, such as... Figure 5 As shown, it includes: The data preprocessing module 510 is used to map the raw data of the vibration response of the wind turbine bearing acquired in real time to a unified spatiotemporal grid using preprocessing methods to obtain a preprocessed spatiotemporal data matrix. The preprocessing methods include spatiotemporal discretization, multi-source data synchronization, noise reduction filtering, normalization, and bilinear interpolation. The fault inversion module 520 is used to perform fault parameter inversion optimization based on the preprocessed spatiotemporal data matrix and the trained inverse physical information neural network model, using the quasi-Newton iteration method with momentum, to obtain the optimal fault parameters of the wind turbine bearing. The result output module 530 is used to analyze the optimal fault parameters using a physical mapping evaluation method to obtain the fault diagnosis results of the wind turbine bearing. The fault diagnosis results of the wind turbine bearing include: quantitative identification results of the depth, width, and spatial location of the wind turbine bearing defects, and qualitative identification results of the fault type.

[0105] In some alternative implementations, the data preprocessing module 510 includes: The time synchronization unit is used to calculate the cross-correlation function between each channel signal and the reference channel signal based on the real-time acquired raw data of the wind turbine bearing vibration response, and extract the time delay corresponding to the peak value to obtain the time-synchronized multi-source vibration data. The standardization processing unit is used to obtain standardized vibration data based on the time-synchronized multi-source vibration data by using a combination of variational mode decomposition denoising, bandpass filtering feature extraction and amplitude normalization. The data interpolation unit is used to map discrete measurement point data to each node of the spatiotemporal grid based on the standardized vibration data and in combination with preset spatiotemporal grid parameters, using bilinear interpolation to obtain a preprocessed spatiotemporal data matrix.

[0106] In some alternative implementations, the fault inversion module 520 includes: The variable determination unit is used to fix all weight matrices and bias vectors for the optimal network parameters of the trained inverse physical information neural network model, and take the set of wind turbine bearing fault features as the variables to be optimized. The parameter construction unit is used to input the preprocessed spatiotemporal data matrix into the inverse physical information neural network model, and to construct the fault parameter inversion optimization function by minimizing the multi-fidelity total loss function as the optimization objective. The iterative solution unit is used to iterate the set of wind turbine bearing fault features based on the fault parameter inversion optimization function and the quasi-Newton iterative method of driving momentum until the iteration termination condition is met; the iteration termination condition includes the number of iterations reaching the preset total number of iterations or the difference between the total loss function values ​​of two adjacent iterations being less than the preset convergence threshold. The parameter output unit is used to determine the optimal fault parameters of the wind turbine bearing based on the iterative set of wind turbine bearing fault characteristics.

[0107] In some alternative implementations, the multifidelity total loss function in the iterative solution unit includes: Based on the predicted vibration response output by the inverse physical information neural network model and the actual observation data of the spatiotemporal data matrix, the data residual term, physical residual term and boundary residual term are obtained by using the residual calculation method. Based on the data residuals, physical residuals, and boundary residuals, the neural tangent kernel algorithm is used to analyze the gradient norm change trend of the corresponding residuals during backpropagation, and the weight coefficients of the corresponding residuals in the total loss function are adjusted to obtain the multi-fidelity total loss function.

[0108] In some alternative implementations, the multifidelity total loss function includes: , , , , in, Represents physical residuals; Indicates data residuals; Indicates boundary residuals; The weighting coefficients represent the physical residuals; The weighting coefficients represent the data residuals; The weighting coefficients representing the boundary residuals; This represents the number of random sampling points used to calculate the physical residual; This represents the bearing seat displacement response predicted by the inverse physical information neural network; Indicates the density of the bearing housing material; This represents the second-order partial derivative of the bearing seat displacement response predicted by the inverse physics information neural network model with respect to time t. Represents the stress divergence term; For fault impact source terms, used to represent the fault impact source term in the first... Spatial location related to each fault and time The magnitude of the impact force at the point; Indicates the number of observed data points; This represents the vibration response predicted by the inverse physics information neural network model at the observation location. and time The value at; This represents the actual observed vibration response; Indicates the number of boundary condition points; This represents the vibration response predicted by the inverse physics information neural network model at the boundary location. and time The value at; This indicates the vibration response specified by the boundary conditions.

[0109] In some alternative implementations, the result output module 530 includes: Based on the optimal fault parameters, quantitative feature parameters are extracted from the optimal fault parameters to obtain quantitative identification results of the defect geometric dimensions and fault spatial location of the wind turbine bearing, and qualitative identification results of the fault type. Based on the quantitative identification results of the defect geometry and fault spatial location, and the qualitative identification results of the fault type, combined with the preset fault severity grading standard, the overall severity of the fault is comprehensively evaluated, and the fault diagnosis results and fault severity grading results of the wind turbine bearing are output.

[0110] In some optional implementations, the optimal fault parameters include: defect depth, defect opening width, fault impact amplitude, defect center coordinates, and optimal characteristic frequency; the step of extracting quantitative characteristic parameters from the optimal fault parameters to obtain quantitative identification results of the defect geometry and fault spatial location of the wind turbine bearing, and qualitative identification results of the fault type, based on the optimal fault parameters, includes: Based on the defect depth and defect opening width, the geometric dimensions of the wind turbine bearing defect are used as quantitative identification results, and the fault impact amplitude is used as an auxiliary quantitative indicator of the fault impact intensity. Based on the coordinates of the defect center, combined with the coordinate system of the radial plane of the wind turbine bearing housing, the spatial position of the defect on the bearing component is analyzed, the specific component and circumferential orientation of the fault are determined, and a quantitative identification result of the fault spatial position is obtained. Based on the optimal characteristic frequency, combined with the bearing geometric kinematics formula, and substituting the inherent structural parameters and real-time operating parameters of the wind turbine bearing, the matching degree between the optimal characteristic frequency and the theoretical value of the characteristic frequency for different fault types of the bearing is verified, and the qualitative identification result of the fault type is determined.

[0111] The wind turbine bearing fault diagnosis device provided in this embodiment of the invention can execute the wind turbine bearing fault diagnosis method provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects for executing the method. Further functional descriptions of the above modules and units are the same as in the corresponding embodiments described above, and will not be repeated here.

[0112] Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention.

[0113] The following is a detailed reference. Figure 6 This diagram illustrates a suitable structural design for implementing an electronic device according to embodiments of the present invention. The electronic device may include a processor (e.g., a central processing unit, graphics processor, etc.) 601, which can perform various appropriate actions and processes based on a program stored in read-only memory (ROM) 602 or a program loaded from memory 608 into random access memory (RAM) 603. RAM 603 also stores various programs and data required for the operation of the electronic device. The processor 601, ROM 602, and RAM 603 are interconnected via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0114] Typically, the following devices can be connected to I / O interface 605: input devices 606 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 607 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; memory devices 608 including, for example, magnetic tapes, hard disks, etc.; and communication devices 609. Communication device 609 allows electronic devices to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 6 Electronic devices with various devices are shown, but it should be understood that it is not required to implement or have all of the devices shown, and more or fewer devices may be implemented or have instead.

[0115] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 609, or installed from a memory 608, or installed from a ROM 602. When the computer program is executed by the processor 601, it performs the functions defined in the wind turbine bearing fault diagnosis method of the embodiments of the present invention.

[0116] Figure 6 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.

[0117] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as computer code that can be recorded on a storage medium, or implemented as computer code downloaded via a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and then stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code. When the software or computer code is accessed and executed by the computer, processor, or hardware, the wind turbine bearing fault diagnosis method shown in the above embodiments is implemented.

[0118] A portion of this invention can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to the invention through the operation of the computer. Those skilled in the art will understand that the forms in which computer program instructions exist in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executing the instructions, or the computer compiling the instructions and then executing the corresponding compiled program, or the computer reading and executing the instructions, or the computer reading and installing the instructions and then executing the corresponding installed program. Here, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to a computer.

[0119] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A method for diagnosing wind turbine bearing faults, characterized in that, The method includes: Based on the raw data of the vibration response of the wind turbine bearing acquired in real time, a preprocessing method is used to map it to a unified spatiotemporal grid to obtain a preprocessed spatiotemporal data matrix. The preprocessing method includes spatiotemporal discretization, multi-source data synchronization, noise reduction filtering, normalization and bilinear interpolation. Based on the preprocessed spatiotemporal data matrix, combined with the trained inverse physical information neural network model, the fault parameters are inverted and optimized using the quasi-Newton iteration method with kinetic momentum to obtain the optimal fault parameters of the wind turbine bearing. Based on the optimal fault parameters, the physical mapping evaluation method is used for analysis to obtain the fault diagnosis results of the wind turbine bearing. The fault diagnosis results of the wind turbine bearing include: quantitative identification results of the depth, width, and spatial location of the wind turbine bearing defects, and qualitative identification results of the fault type.

2. The method according to claim 1, characterized in that, Based on the preprocessed spatiotemporal data matrix, combined with the trained inverse physical information neural network model, the fault parameters are inverted and optimized using the quasi-Newton iterative method with driving forces to obtain the optimal fault parameters of the wind turbine bearing, including: To optimize the network parameters of the trained inverse physical information neural network model, all weight matrices and bias vectors are fixed, and the set of wind turbine bearing fault features is taken as the variable to be optimized. Based on the preprocessed spatiotemporal data matrix, it is input into the inverse physical information neural network model, and the multi-fidelity total loss function is minimized as the optimization objective to construct the fault parameter inversion optimization function. Based on the fault parameter inversion optimization function, the quasi-Newton iteration method with driving force is used to iterate the set of fault features of wind turbine bearings until the iteration termination condition is met; the iteration termination condition includes the number of iterations reaching the preset total number of iterations or the difference between the total loss function values ​​of two adjacent iterations being less than the preset convergence threshold. Based on the iterative set of wind turbine bearing fault characteristics, the optimal fault parameters of the wind turbine bearing are determined.

3. The method according to claim 2, characterized in that, The multi-fidelity total loss function includes: Based on the predicted vibration response output by the inverse physical information neural network model and the actual observation data of the spatiotemporal data matrix, the data residual term, physical residual term and boundary residual term are obtained by using the residual calculation method. Based on the data residuals, physical residuals, and boundary residuals, the neural tangent kernel algorithm is used to analyze the gradient norm change trend of the corresponding residuals during backpropagation, and the weight coefficients of the corresponding residuals in the total loss function are adjusted to obtain the multi-fidelity total loss function.

4. The method according to claim 3, characterized in that, The multi-fidelity total loss function includes: ， ， ， ， in, Represents physical residuals; Indicates data residuals; Indicates boundary residuals; The weighting coefficients represent the physical residuals; The weighting coefficients represent the data residuals; The weighting coefficients representing the boundary residuals; This represents the number of random sampling points used to calculate the physical residual; This represents the bearing seat displacement response predicted by the inverse physical information neural network; Indicates the density of the bearing housing material; This represents the second-order partial derivative of the bearing seat displacement response predicted by the inverse physics information neural network model with respect to time t. Represents the stress divergence term; For fault impact source terms, used to represent the fault impact source term in the first... Spatial location related to each fault and time The magnitude of the impact force at the point; Indicates the number of observed data points; This represents the vibration response predicted by the inverse physics information neural network model at the observation location. and time The value at; This represents the actual observed vibration response; Indicates the number of boundary condition points; This represents the vibration response predicted by the inverse physics information neural network model at the boundary location. and time The value at; This indicates the vibration response specified by the boundary conditions.

5. The method according to claim 1, characterized in that, The raw data of wind turbine bearing vibration response acquired in real time is preprocessed and mapped to a unified spatiotemporal grid to obtain a preprocessed spatiotemporal data matrix, including: Based on the real-time acquired raw data of wind turbine bearing vibration response, the cross-correlation function between each channel signal and the reference channel signal is calculated using the cross-correlation method, and the time delay corresponding to the peak value is extracted to obtain multi-source vibration data after time synchronization. Based on the time-synchronized multi-source vibration data, standardized vibration data is obtained by using a combination of variational mode decomposition denoising, bandpass filtering feature extraction, and amplitude normalization. Based on the standardized vibration data and combined with the preset spatiotemporal grid parameters, the discrete measurement point data is mapped to each node of the spatiotemporal grid using bilinear interpolation to obtain the preprocessed spatiotemporal data matrix.

6. The method according to claim 1, characterized in that, The fault diagnosis results of the wind turbine bearing are obtained by analyzing the optimal fault parameters using a physical mapping evaluation method, including: Based on the optimal fault parameters, quantitative feature parameters are extracted from the optimal fault parameters to obtain quantitative identification results of the defect geometric dimensions and fault spatial location of the wind turbine bearing, and qualitative identification results of the fault type. Based on the quantitative identification results of the defect geometry and fault spatial location, and the qualitative identification results of the fault type, combined with the preset fault severity grading standard, the overall severity of the fault is comprehensively evaluated, and the fault diagnosis results and fault severity grading results of the wind turbine bearing are output.

7. The method according to claim 6, characterized in that, The optimal fault parameters include: defect depth, defect opening width, fault impact amplitude, defect center coordinates, and optimal characteristic frequency. Based on these optimal fault parameters, quantitative characteristic parameters are extracted to obtain quantitative identification results of the defect geometry and fault spatial location of the wind turbine bearing, and qualitative identification results of the fault type, including: Based on the defect depth and defect opening width, the geometric dimensions of the wind turbine bearing defect are used as quantitative identification results, and the fault impact amplitude is used as an auxiliary quantitative indicator of the fault impact intensity. Based on the coordinates of the defect center, combined with the coordinate system of the radial plane of the wind turbine bearing housing, the spatial position of the defect on the bearing component is analyzed, the specific component and circumferential orientation of the fault are determined, and a quantitative identification result of the fault spatial position is obtained. Based on the optimal characteristic frequency, combined with the bearing geometric kinematics formula, and substituting the inherent structural parameters and real-time operating parameters of the wind turbine bearing, the matching degree between the optimal characteristic frequency and the theoretical value of the characteristic frequency for different fault types of the bearing is verified, and the qualitative identification result of the fault type is determined.

8. A wind turbine bearing fault diagnosis device, characterized in that, The device includes: The data preprocessing module is used to map the raw data of the vibration response of the wind turbine bearing acquired in real time to a unified spatiotemporal grid using preprocessing methods to obtain a preprocessed spatiotemporal data matrix. The preprocessing methods include spatiotemporal discretization, multi-source data synchronization, noise reduction filtering, normalization, and bilinear interpolation. The fault inversion module is used to perform fault parameter inversion optimization based on the preprocessed spatiotemporal data matrix and the trained inverse physical information neural network model, using the quasi-Newton iteration method with momentum, to obtain the optimal fault parameters of the wind turbine bearing. The result output module is used to analyze the optimal fault parameters using a physical mapping evaluation method to obtain the fault diagnosis results of the wind turbine bearing. The fault diagnosis results of the wind turbine bearing include: quantitative identification results of the depth, width, and spatial location of the wind turbine bearing defects, and qualitative identification results of the fault type.

9. An electronic device, characterized in that, include: The system includes a memory and a processor, which are interconnected and the memory stores computer instructions. The processor executes the computer instructions to perform the wind turbine bearing fault diagnosis method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing the computer to execute the wind turbine bearing fault diagnosis method according to any one of claims 1 to 7.

Images