Fan blade damage identification method, device and equipment and medium

Abstract

The invention provides a fan blade damage identification method and device, equipment and a medium, and relates to the technical field of wind power fault diagnosis, and the method comprises the steps: carrying out the self-adaptive mode decomposition, denoising and reconstruction processing of original strain signal data, and obtaining a reconstruction signal; dividing the reconstructed signal into a plurality of signal segments, taking each signal segment as a graph node, determining the similarity between the graph nodes according to the signal features corresponding to the graph nodes, determining the neighbor nodes for each graph node based on the similarity, constructing the edge between each graph node and the neighbor nodes thereof based on the similarity, and obtaining the graph node. Obtaining graph structure data based on each graph node and edge; and determining the damage position and / or damage severity of the fan blade corresponding to the graph structure data based on the trained damage identification graph neural network model so as to improve the accuracy of damage identification and realize the position positioning and severity evaluation of single damage and multiple damages at the same time.

Description

Technical Field

[0001] This invention relates to the field of wind power fault diagnosis technology, and more specifically, to a method, apparatus, equipment, and medium for identifying damage to wind turbine blades. Background Technology

[0002] Existing methods for identifying damage to offshore wind turbine blades are mainly divided into machine vision methods and response signal methods. In terms of signal processing and damage identification algorithms, traditional machine learning algorithms (such as support vector machines and K-nearest neighbors) suffer from low accuracy and poor generalization ability. Existing deep learning algorithms (such as neural networks and long short-term memory networks) have weak modeling ability when processing damage data of blades with non-Euclidean structures, resulting in large errors in damage localization and severity identification. Furthermore, the accuracy of damage identification drops significantly when faced with strong noise interference in the marine environment. Summary of the Invention

[0003] In view of this, the purpose of the present invention is to provide a method, device, equipment and medium for identifying damage to wind turbine blades, so as to improve the accuracy of damage identification and realize the location and severity assessment of single and multiple damages.

[0004] In a first aspect, this application provides a method for identifying damage to wind turbine blades, including: Acquire raw strain signal data of the wind turbine blades; The original strain signal data is subjected to adaptive mode decomposition, denoising, and reconstruction to obtain the reconstructed signal. The reconstructed signal is divided into multiple signal segments, and each signal segment is treated as a graph node. Based on the signal features corresponding to the graph nodes, the similarity between each graph node is determined. Based on the similarity, the neighboring nodes of each graph node are determined. Based on the similarity, the edges between each graph node and its neighboring nodes are constructed. Based on each graph node and edge, the graph structure data is obtained. The damage identification result of the wind turbine blade corresponding to the graph structure data is determined based on the trained damage identification graph neural network model; wherein, the damage identification result includes the damage location and / or damage severity.

[0005] Optionally, the original strain signal data undergoes adaptive mode decomposition, denoising, and reconstruction to obtain a reconstructed signal, including: A metaheuristic optimization algorithm is used to adaptively optimize the parameters of successive variational mode decomposition to obtain the optimized parameters; Based on the optimized parameters, the original strain signal data is decomposed by successive variational mode decomposition to obtain multiple intrinsic mode function components. Multiple intrinsic mode function components are filtered and denoised to obtain the denoised component signals. The denoised component signals are superimposed to obtain the reconstructed signal.

[0006] Optionally, a metaheuristic optimization algorithm is used to adaptively optimize the parameters of the successive variational mode decomposition, resulting in optimized parameters including: The penalty factor parameters of successive variational mode decomposition are adaptively optimized using the walrus optimization algorithm to obtain optimized penalty factor parameters; wherein, the optimized penalty factor parameters are the minimum of the envelope entropy of the intrinsic mode function components after decomposition.

[0007] Optionally, the reconstructed signal is divided into multiple signal segments, including: The reconstructed signal is divided into multiple signal segments using a non-overlapping sliding window.

[0008] Optionally, based on similarity, each graph node determines its neighboring nodes, including: Based on the similarity between a graph node and other graph nodes, the K-nearest neighbor algorithm is used to determine its set of neighbor nodes.

[0009] Optionally, the edges between each graph node and its neighboring nodes are determined based on similarity, including: Based on each neighbor node in the set of neighbor nodes, the edge weights between the graph node and its neighbors are determined using a Gaussian kernel function according to the similarity between the graph node and its neighbors. Based on edge weights, the edges between graph nodes and their neighboring nodes are determined.

[0010] Optionally, the wind turbine blade damage identification method provided in this application further includes: Obtain a training dataset; the training dataset includes multiple training sample data; each training sample data includes graph structure training data and corresponding damage location label data and damage severity label data; the graph structure training data is obtained by adaptive modal decomposition, denoising and reconstruction processing of the original strain signal data to obtain the reconstructed signal; the reconstructed signal is divided into multiple signal segments, and each signal segment is taken as a graph node. Based on the signal features corresponding to the graph nodes, the similarity between each graph node is determined. Based on the similarity, the neighboring nodes of each graph node are determined. Based on the similarity, the edges between each graph node and its neighboring nodes are constructed. Based on each graph node and edge, the following is obtained: Based on the training dataset, iterative training is performed on the initial damage recognition graph neural network model until the termination condition is met. Then, based on the parameters of the initial damage recognition graph neural network model updated during the last iteration, a new damage recognition graph neural network model is obtained. The iterative training operations include: Select target training sample data from the training dataset; Input the graph structure training data from the target training sample data into the initial damage recognition graph neural network model to obtain the classification prediction results of the damage location and the regression prediction results of the damage severity. The cross-entropy loss is determined based on the classification prediction results and the damage location label data in the target training sample data, and the mean squared error loss is determined based on the regression prediction results and the damage severity label data in the target training sample data; the parameters of the initial damage recognition graph neural network model are updated based on the cross-entropy loss and the mean squared error loss.

[0011] Secondly, this application provides a wind turbine blade damage identification device, comprising: The data acquisition module is used to acquire the raw strain signal data of the wind turbine blades; The data preprocessing module is used to perform adaptive mode decomposition, denoising, and reconstruction on the raw strain signal data to obtain the reconstructed signal; The graph structure data module is used to divide the reconstructed signal into multiple signal segments and treat each signal segment as a graph node. Based on the signal characteristics corresponding to the graph node, the similarity between each graph node is determined. Based on the similarity, the neighboring nodes of each graph node are determined. Based on the similarity, the edges between each graph node and its neighboring nodes are constructed. Based on each graph node and edge, the graph structure data is obtained. The damage identification module is used to determine the damage identification result of the wind turbine blade corresponding to the graph structure data based on the trained damage identification graph neural network model; wherein, the damage identification result includes the damage location and / or damage severity.

[0012] Thirdly, this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-mentioned wind turbine blade damage identification method.

[0013] Fourthly, this application provides a computer-readable storage medium storing computer instructions, which, when executed by a processor, implement the aforementioned wind turbine blade damage identification method.

[0014] This invention provides a method, apparatus, device, and medium for identifying wind turbine blade damage. The method involves adaptive mode decomposition, denoising, and reconstruction of the original strain signal data to obtain a reconstructed signal. This reconstructed signal is then divided into multiple signal segments, each representing a graph node. Based on the signal characteristics corresponding to each graph node, the similarity between the nodes is determined. Based on this similarity, neighboring nodes are identified for each node. Edges between each node and its neighbors are constructed, and graph structure data is obtained based on the graph nodes and edges. Finally, a trained damage identification graph neural network model is used to determine the damage location and / or severity of the wind turbine blade corresponding to the graph structure data. This improves the accuracy of damage identification while simultaneously enabling location and severity assessment of single and multiple damages.

[0015] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 A flowchart of a wind turbine blade damage identification method provided by an embodiment of the present invention is shown; Figure 2 A flowchart of the damage recognition graph neural network model training method provided in an embodiment of the present invention is shown; Figure 3 This diagram illustrates the structure of a wind turbine blade damage identification device provided in an embodiment of the present invention. Figure 4 A schematic diagram of the structure of an electronic device provided in an embodiment of the present invention is shown. Detailed Implementation

[0018] 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, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0019] This application provides a method for identifying wind turbine blade damage. (See attached document.) Figure 1 As shown in the embodiments of this application, the wind turbine blade damage identification method includes: Step 110: Obtain the original strain signal data of the wind turbine blades.

[0020] In this embodiment of the application, the raw strain signal data of the wind turbine blades can be obtained in the following ways: Multiple fiber Bragg grating sensors distributed on the surface of the wind turbine blades collect strain response signals of the blades under operating loads as raw strain signal data.

[0021] Step 120: Perform adaptive mode decomposition, denoising, and reconstruction on the original strain signal data to obtain the reconstructed signal.

[0022] In this embodiment, a metaheuristic optimization algorithm is used to adaptively optimize the parameters of successive variational mode decomposition (SMD) to obtain optimized parameters. Based on the optimized parameters, SMD is used to decompose the original strain signal data to obtain multiple intrinsic mode function (IMF) components. These IMF components are then filtered and denoised to obtain denoised component signals. Finally, the denoised component signals are superimposed to obtain a reconstructed signal. Furthermore, a walrus optimization algorithm is used to adaptively optimize the penalty factor parameters of SMD to obtain optimized penalty factor parameters. The optimized penalty factor parameters minimize the envelope entropy of the decomposed IMF components.

[0023] Since raw strain signal data typically contains a large amount of noise interference, directly processing it using traditional variational mode decomposition methods can easily lead to unsatisfactory signal decomposition results, thus affecting the accurate extraction of damage features. Therefore, this application proposes using a walrus optimization algorithm to adaptively optimize the penalty factor parameters of successive variational mode decomposition, improving the accuracy and stability of signal decomposition while overcoming the challenges of high signal noise interference and blurred damage features in the marine environment.

[0024] In practical implementation, the Walrus Optimization Algorithm (WaOA) is used to simulate the foraging, migration, predator avoidance, and predator defense behaviors of walruses. The optimal penalty factor parameter for Successive Variational Mode Decomposition (SVMD) is searched through four stages: initialization, exploration, migration, and development. The specific process is as follows: Step 1: Initialization phase, treating each walrus as a penalty factor parameter. Candidate solutions are used to construct a population matrix. ,in, , For the number of walruses, As a parameter, in this application, the number of walruses Optimized parameters (Only optimize the penalty factor parameters) ), For the first The penalty factor parameter corresponding to each walrus The candidate values ​​are set to a range of values. The objective function is determined to be minimizing the envelope entropy of the decomposed signal, where the envelope entropy of the decomposed signal can be determined by the following formula:

[0025] In the formula, For the envelope entropy, The sum of the number of intrinsic mode functions obtained after mode decomposition of the signal. The amplitude of the demodulated envelope spectrum of the intrinsic mode functions obtained after mode decomposition of the signal is given. This is a normalized probability distribution; Step 2: Exploration Phase, Generating New Locations If the objective function value at the new position is better, then the original position is replaced. The objective function value at the new position can be obtained using the following formula:

[0026] In the formula, To explore new locations Values ​​on the dimension, For the first The penalty factor parameter corresponding to each walrus Candidate values, It is a random number. The optimal walrus individual in the current population (i.e., the individual with the optimal objective function value) at the [number]th [time] Values ​​on the dimension, It is a random integer; If the current individual is in a new location Corresponding objective function value Less than the The objective function value (i.e., envelope entropy value) corresponding to the current position of each walrus individual. ,but If it is not less than, then retain its original position; Step 3: The migration phase, moving to other randomly selected walruses. Directional migration generates new locations The new location generated during migration can be determined using the following formula:

[0027] In the formula, The first to generate new locations for migration Values ​​on the dimension, For the first The first walrus individual in the The values ​​that can be taken on the dimension parameter, For the first The objective function value for an individual walrus; If migration generates a new location Less than the The objective function value corresponding to the current position of each walrus individual ,but If it is not less than, then retain its original position; Step 4: Development phase, searching for new locations within the neighborhood of the original candidate solutions. The neighborhood radius changes with iteration, and the discovery of new locations can be determined by the following formula:

[0028] In the formula, t is the number of iterations. , These are the local upper and lower limits for the j-th variable; If a new location is developed Less than the The objective function value corresponding to the current position of each walrus individual ,but If it is not less than, then retain its original position; In this application, after 100 iterations, the optimal penalty factor parameter is output. value.

[0029] Furthermore, based on the optimized penalty factor parameters This application employs SVMD to decompose the original strain signal into multiple intrinsic mode function components (IMFs). SVMD achieves adaptive mode decomposition by constructing a constrained optimization model, without requiring a preset number of modes. For example, assuming the input signal... Decomposed into L-order modes and residual signal (including 1 to (the sum of the first and second modes), that is: (Unprocessed signal) Among them, several constraints are: The first constraint: each mode revolves around the center frequency The focus is on minimizing the constraint (i.e., the first constraint). for:

[0030] In the formula, For time partial derivatives, For the Dirac function, The imaginary unit, For continuous time variables, For convolution, It is a complex exponential function. For the first The center angular frequency of the first-order eigenmode function; Second constraint: Minimize and Spectral overlap, impulse response through mode filter frequency response Establish a second constraint ; Second constraint: For The first mode adopts a similar The treatment of first-order modes establishes a third constraint. ; Based on the above constraints, an optimization problem for solving the augmented Lagrangian function is constructed, that is, each IMF is obtained iteratively by the alternating direction multiplier method. Furthermore, a dual-threshold strategy is employed to screen effective IMFs. In this application, the correlation coefficient between each IMF and the original strain signal data is calculated. :

[0031] In the formula, The signal sampling length, For the first One IMF; During the denoising process, the correlation coefficients are sorted in ascending order, and the 80th percentile is taken as the high threshold. The 50th percentile is the low threshold. ;like If the IMFs are the dominant components of the signal, meaning they contain a large number of effective features, then SG (Savitzky-Golay) filtering can be used for smoothing and noise reduction; if Then the IMFs are a mixture of signal and noise components, which can be denoised using wavelet thresholding; if If the IMFs are noise-dominated components, they should be discarded directly. During the reconstruction process, the effective IMFs after SG filtering and wavelet threshold denoising are superimposed to obtain the denoised reconstructed signal, which is used for subsequent damage identification.

[0032] It should be noted that IMFs are the complex form of IMFs, i.e., multiple intrinsic mode functions. In practical signal decomposition, the original signal is usually decomposed into a series (multiple) IMFs, each IMF representing an oscillation mode within a specific frequency range of the signal; the 80th percentile is the value at the 80th percentile after sorting all the correlation coefficients between the intrinsic mode functions and the original signal in ascending order, and the value at the 50th percentile after sorting is taken as the high threshold; the 50th percentile is the value at the 50th percentile after sorting all the correlation coefficients in ascending order, and the value at the 50th percentile after sorting is taken as the low threshold.

[0033] Step 130: Divide the reconstructed signal into multiple signal segments and treat each signal segment as a graph node. Determine the similarity between each graph node based on the signal features corresponding to the graph node. Based on the similarity, determine the neighboring nodes of each graph node. Based on the similarity, construct the edges between each graph node and its neighboring nodes. Based on each graph node and edge, obtain the graph structure data.

[0034] In this embodiment of the application, graph structure data can be obtained in the following ways: First, the reconstructed signal is divided using a non-overlapping sliding window to obtain multiple signal segments. In this application, the reconstructed signal is divided into sub-samples using a non-overlapping sliding window (i.e., the length of the non-overlapping sliding window is 40). Each subsample corresponds to a label. (i.e., injury location / injury severity labels), forming a sample set. ,in ,in, The total length of the reconstructed signal. The length of the non-overlapping sliding window. This represents the total number of subsamples obtained after segmentation. Then, based on the similarity between a graph node and other graph nodes, the K-nearest neighbor algorithm is used to determine its set of neighboring nodes; and based on each neighboring node in the set, a Gaussian kernel function is used to determine the edge weights between the graph node and its neighbors according to the similarity between the graph node and its neighbors; based on the edge weights, the edges between the graph node and its neighbors are determined. In this application, each subsample is considered as a graph node, and for each node... From the candidate set Selected from Nearest neighbor as neighbor node Using the Gaussian kernel function, the node values ​​are calculated using the following formula. with neighboring nodes edge weight :

[0035] In the formula, For nodes With nodes The Euclidean distance between them This refers to the Gaussian kernel bandwidth parameter; Finally, based on each graph node and edge, the graph structure data is obtained.

[0036] In this application, the walrus optimization algorithm is used to adaptively optimize the signal decomposition parameters, thereby improving the accuracy and robustness of the signal preprocessing stage. By using dual threshold screening and multi-level denoising strategies, noise interference in complex environments is effectively suppressed, enhancing the identification of damage features. By converting time-series signals into graph-structured data and using a graph sampling aggregation network for feature learning, the limitations of traditional deep learning methods in processing non-Euclidean data are overcome, achieving accurate and stable identification of the location and severity of blade damage.

[0037] Step 140: Determine the damage identification result of the wind turbine blade corresponding to the graph structure data based on the trained damage identification graph neural network model; wherein, the damage identification result includes the damage location and / or damage severity.

[0038] In this embodiment, based on the trained damage recognition graph neural network model, the input wind turbine blade diagram structure data is processed to obtain the corresponding damage recognition result. The damage recognition result includes the damage location and / or damage severity. The output layer of the trained damage recognition graph neural network model is as follows: if it is a classification task, it outputs the category label corresponding to the damage location; if it is a regression task, it outputs a continuous value representing the damage severity; if it is a multi-task learning structure, it can simultaneously output the location category and severity estimate, which can realize automated and accurate assessment of blade status and provide real-time and reliable basis for operation and maintenance decisions.

[0039] This application provides a method for training a damage recognition graph neural network model, see below. Figure 2 As shown, the damage recognition graph neural network model training method provided in this application embodiment includes: Step 210: Obtain the training data set; wherein, the training data set includes multiple training sample data; each training sample data includes graph structure training data and damage location label data and damage severity label data corresponding to the graph structure training data; the graph structure training data is obtained by adaptive mode decomposition, denoising and reconstruction processing of the original strain signal data to obtain the reconstructed signal; the reconstructed signal is divided into multiple signal segments, and each signal segment is taken as a graph node. According to the signal features corresponding to the graph node, the similarity between each graph node is determined. Based on the similarity, the neighboring nodes of each graph node are determined. Based on the similarity, the edges between each graph node and its neighboring nodes are constructed. Based on each graph node and edge, the following is obtained: Step 220: Select target training sample data from the training dataset; Step 230: Input the graph structure training data from the target training sample data into the initial damage recognition graph neural network model to obtain the classification prediction results of the damage location and the regression prediction results of the damage severity; determine the cross-entropy loss based on the classification prediction results and the damage location label data in the target training sample data, and determine the mean squared error loss based on the regression prediction results and the damage severity label data in the target training sample data; update the parameters of the initial damage recognition graph neural network model based on the cross-entropy loss and the mean squared error loss. Step 240: Determine whether the iterative training termination condition is met; if yes, proceed to step 250; if no, return to step 220; wherein, the iterative training termination condition is that the number of iterations is not less than the number threshold, or the prediction error is not higher than the error threshold.

[0040] Step 250: Based on the parameters of the initial damage recognition graph neural network model updated during the last iteration of training, obtain the damage recognition graph neural network model.

[0041] In this embodiment, the damage recognition graph neural network model includes an input layer, two graph neural network layers, and two fully connected layers. Specifically, the damage recognition graph neural network model includes an input layer, a first graph sampling and aggregation layer, a first batch normalization layer, a first modified linear unit activation function layer, a second graph sampling and aggregation layer, a second batch normalization layer, a second modified linear unit activation function layer, a first fully connected layer, and a second fully connected layer connected in sequence. The graph sampling and aggregation layer converts the input node feature dimension from 40 to 1024 dimensions. The batch normalization layer performs batch normalization processing on the 1024-dimensional node features output by the first graph sampling and aggregation layer. The modified linear unit activation function layer performs a nonlinear transformation on each feature value output by the first batch normalization layer. If the input is greater than 0, the original value is output; otherwise, 0 is output. The first fully connected layer maps the 1024-dimensional node features to 512 dimensions. The second fully connected layer maps the 512-dimensional features to the final output damage location and / or damage severity. Furthermore, the first image sampling and aggregation layer are processed through an aggregation function. Aggregate the features of neighboring nodes and combine them with the features of the current node to update the node embedding, using the following formula:

[0042] In the formula, For nodes The intermediate features obtained after the current aggregation operation are updated. It is a non-linear activation function (ReLU). For the current node feature, Aggregate neighbor node features, For nodes a certain neighbor node feature, For nodes The set of neighboring nodes; After k rounds of aggregation iterations, the node features are updated as follows:

[0043] in, For nodes go through Features obtained after round (layer) aggregation iteration, For activation function, For learnable weight matrix, For nodes initial characteristics, For nodes neighboring nodes In the Features obtained after round aggregation; For fully connected layers, if the injury localization task is treated as a multi-class classification problem, the cross-entropy loss function can be used:

[0044] In the formula, 1 represents the cross-entropy loss function value used in multi-class classification tasks, which measures the difference between the probability distribution predicted by the model and the true label distribution. This represents the number of samples (nodes) in the current training batch. The total number of categories of damage locations. For the first The true label of the sample in the category The value is determined using one-hot encoding, meaning that if the sample belongs to category... The value is 1 if it is not 0 otherwise. The model predicts the first The sample belongs to the category The probability values ​​are usually converted into a probability distribution by the Softmax function; The task of assessing the severity of damage can be viewed as a regression problem, and the mean squared error loss function can be used:

[0045] In the formula, This is the mean squared error loss function value used in regression tasks, which measures the average squared error between the model's predicted values ​​and the true values. For the first The true severity of injury for each sample is a continuous value. The model predicts the first The severity value of damage for each sample; Furthermore, in this application, the Adam optimizer is used for model training, with the initial learning rate set to 0.001, the number of iterations set to 100, and the weight decay set to [missing value]. During training, the learning rate is gradually reduced to improve the convergence stability and accuracy of the model in the later stages of training. The data is divided into a training set and a test set, with the training set accounting for 70% and the test set accounting for 30%, which are used for model training and generalization performance evaluation.

[0046] This application defines the loss functions and their parameter meanings for two tasks: wind turbine blade damage localization (classification) and damage severity assessment (regression). It also clarifies the key hyperparameters and data partitioning methods during model training. Cross-entropy loss is suitable for multi-classification problems, aiming to minimize the difference between predicted probabilities and the true distribution. Mean squared error loss is suitable for regression problems, aiming to make the predicted values ​​as close as possible to the true values. Combined with training strategies such as the Adam optimizer, learning rate decay, and weight decay, the learning efficiency and generalization ability of the model can be effectively improved, ultimately achieving accurate identification and assessment of blade damage.

[0047] This application provides a wind turbine blade damage identification device, see below. Figure 3 As shown, the wind turbine blade damage identification device provided in this application embodiment includes: The data acquisition module 310 is used to acquire the raw strain signal data of the wind turbine blades; The data preprocessing module 320 is used to perform adaptive mode decomposition, denoising and reconstruction processing on the original strain signal data to obtain the reconstructed signal; The graph structure data module 330 is used to divide the reconstructed signal into multiple signal segments and treat each signal segment as a graph node. Based on the signal features corresponding to the graph nodes, the similarity between each graph node is determined. Based on the similarity, the neighboring nodes of each graph node are determined. Based on the similarity, the edges between each graph node and its neighboring nodes are constructed. Based on each graph node and edge, the graph structure data is obtained. The damage identification module 340 is used to determine the damage identification result of the wind turbine blade corresponding to the graph structure data based on the trained damage identification graph neural network model; wherein, the damage identification result includes the damage location and / or damage severity.

[0048] It should be noted that the principle of the wind turbine blade damage identification device provided in this application embodiment to solve the technical problem is similar to that of the wind turbine blade damage identification method provided in this application embodiment. Therefore, the implementation of the wind turbine blade damage identification device provided in this application embodiment can refer to the implementation of the wind turbine blade damage identification method provided in this application embodiment, and the repeated parts will not be described again.

[0049] After introducing the wind turbine blade damage identification method and device provided in the embodiments of this application, the electronic equipment provided in the embodiments of this application will be briefly introduced next.

[0050] See Figure 4 As shown, the electronic device 500 provided in this application embodiment includes at least a processor 501, a memory 502, and a computer program stored in the memory 502 and executable on the processor 501. When the processor 501 executes the computer program, it implements the wind turbine blade damage identification method provided in this application embodiment.

[0051] The electronic device 500 provided in this application embodiment may further include a bus 503 connecting different components (including processor 501 and memory 502). The bus 503 represents one or more types of bus structures, including memory bus, peripheral bus, local area bus, etc.

[0052] Memory 502 may include a readable storage medium in the form of volatile memory, such as random access memory (RAM) 5021 and / or cache memory 5022, and may further include read-only memory (ROM) 5023. Memory 502 may also include a program tool 5025 having a set (at least one) of program modules 5024, including but not limited to an operating subsystem, one or more application programs, other program modules, and program data, each or some combination of these examples may include an implementation of a network environment.

[0053] Processor 501 can be a single processing element or a collective term for multiple processing elements. For example, processor 501 can be a central processing unit (CPU) or one or more integrated circuits configured to implement the wind turbine blade damage identification method provided in the embodiments of this application. Specifically, processor 501 can be a general-purpose processor, including but not limited to CPUs, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.

[0054] Electronic device 500 can communicate with one or more external devices 504 (e.g., keyboard, remote control, etc.), and also with one or more devices that enable a user to interact with electronic device 500 (e.g., mobile phone, computer, etc.), and / or with devices that enable electronic device 500 to communicate with one or more other electronic devices 500 (e.g., router, modem, etc.). This communication can be performed through input / output (I / O) interface 505. Furthermore, electronic device 500 can also communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN), and / or public networks, such as the Internet) through network adapter 506. Figure 4As shown, network adapter 506 communicates with other modules of electronic device 500 via bus 503. It should be understood that, although... Figure 4 As not shown, other hardware and / or software modules may be used in conjunction with the electronic device 500, including but not limited to microcode, device drivers, redundant processors, external disk drive arrays, Redundant Arrays of Independent Disks (RAID) subsystems, tape drives, and data backup storage subsystems.

[0055] It should be noted that, Figure 4 The electronic device 500 shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0056] The computer-readable storage medium provided in the embodiments of this application is described below. The computer-readable storage medium provided in the embodiments of this application stores computer instructions, which, when executed by a processor, implement the wind turbine blade damage identification method provided in the embodiments of this application. Specifically, the computer instructions can be built into or installed in the processor, so that the processor can implement the wind turbine blade damage identification method provided in the embodiments of this application by executing the built-in or installed computer instructions.

[0057] In addition, the wind turbine blade damage identification method provided in this application embodiment can also be implemented as a computer program product, which includes program code. The program code implements the wind turbine blade damage identification method provided in this application embodiment when it is run on a processor.

[0058] The computer program product provided in this application embodiment may employ one or more computer-readable storage media, which may be, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination thereof. Specifically, more specific examples (a non-exhaustive list) of computer-readable storage media include electrical connections with one or more wires, portable disks, hard disks, RAM, ROM, erasable programmable read-only memory (EPROM), optical fibers, portable compact disc read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0059] The computer program product provided in this application embodiment can be a CD-ROM and include program code, and can also run on electronic devices such as computers. However, the computer program product provided in this application embodiment is not limited thereto. In this application embodiment, the computer-readable storage medium can be any tangible medium that contains or stores program code, which can be used by or in conjunction with an instruction execution system, device, or apparatus.

[0060] It should be noted that although several units or sub-units of the device have been mentioned in the detailed description above, this division is merely exemplary and not mandatory. In fact, according to embodiments of this application, the features and functions of two or more units described above can be embodied in one unit. Conversely, the features and functions of one unit described above can be further divided and embodied by multiple units.

[0061] Furthermore, although the operations of the method of this application are described in a specific order in the accompanying drawings, this does not require or imply that these operations must be performed in that specific order, or that all the operations shown must be performed to achieve the desired result. Additionally or alternatively, certain steps may be omitted, multiple steps may be combined into one step, and / or one step may be broken down into multiple steps.

[0062] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0063] Obviously, those skilled in the art can make various modifications and variations to the embodiments of this application without departing from the spirit and scope of the embodiments of this application. Therefore, if these modifications and variations to the embodiments of this application fall within the scope of the claims of this application and their equivalents, this application also intends to include these modifications and variations.

Claims

1. A method for identifying damage to wind turbine blades, characterized in that, include: Acquire raw strain signal data of the wind turbine blades; The original strain signal data is subjected to adaptive mode decomposition, denoising, and reconstruction to obtain the reconstructed signal; The reconstructed signal is divided into multiple signal segments, and each signal segment is treated as a graph node. The similarity between the graph nodes is determined based on the signal features corresponding to the graph nodes. Based on the similarity, the neighboring nodes of each graph node are determined. Based on the similarity, the edges between each graph node and its neighboring nodes are constructed. Based on each graph node and the edges, graph structure data is obtained. The damage identification result of the wind turbine blade corresponding to the graph structure data is determined based on the trained damage identification graph neural network model; wherein, the damage identification result includes the damage location and / or damage severity.

2. The wind turbine blade damage identification method according to claim 1, characterized in that, The original strain signal data is subjected to adaptive mode decomposition, denoising, and reconstruction processing to obtain a reconstructed signal, including: A metaheuristic optimization algorithm is used to adaptively optimize the parameters of successive variational mode decomposition to obtain the optimized parameters; Based on the optimized parameters, the original strain signal data is decomposed by successive variational mode decomposition to obtain multiple intrinsic mode function components. The intrinsic mode function components are filtered and denoised to obtain the denoised component signals. The denoised component signals are superimposed to obtain the reconstructed signal.

3. The wind turbine blade damage identification method according to claim 2, characterized in that, A metaheuristic optimization algorithm is used to adaptively optimize the parameters of the successive variational mode decomposition, resulting in the following optimized parameters: The penalty factor parameters of the successive variational mode decomposition are adaptively optimized using the walrus optimization algorithm to obtain optimized penalty factor parameters; wherein, the optimized penalty factor parameters are the minimum of the envelope entropy of the intrinsic mode function components after decomposition.

4. The wind turbine blade damage identification method according to claim 1, characterized in that, The reconstructed signal is divided into multiple signal segments, including: The reconstructed signal is divided using a non-overlapping sliding window to obtain multiple signal segments.

5. The wind turbine blade damage identification method according to claim 4, characterized in that, The step of determining the neighbor nodes for each graph node based on the similarity includes: Based on the similarity between the graph node and the other graph nodes, the K-nearest neighbor algorithm is used to determine its set of neighbor nodes.

6. The wind turbine blade damage identification method according to claim 5, characterized in that, Determining the edges between each graph node and its neighboring nodes based on the similarity includes: Based on each neighbor node in the set of neighbor nodes, the edge weights between the graph node and its neighbor nodes are determined using a Gaussian kernel function according to the similarity between the graph node and its neighbor nodes. Based on the edge weights, the edges between the graph node and its neighboring nodes are determined.

7. The wind turbine blade damage identification method according to claim 1, characterized in that, Also includes: A training data set is obtained; wherein the training data set includes multiple training sample data; each training sample data includes graph structure training data and damage location label data and damage severity label data corresponding to the graph structure training data; the graph structure training data is obtained by performing adaptive mode decomposition, denoising and reconstruction processing on the original strain signal data to obtain a reconstructed signal; the reconstructed signal is divided into multiple signal segments, and each signal segment is taken as a graph node; the similarity between each graph node is determined according to the signal features corresponding to the graph node; based on the similarity, its neighboring nodes are determined for each graph node; based on the similarity, the edges between each graph node and its neighboring nodes are constructed; based on each graph node and the edges, the following is obtained: Based on the training dataset, an iterative training operation is performed on the initial damage recognition graph neural network model until the iterative training termination condition is met. Then, based on the parameters of the initial damage recognition graph neural network model updated during the last execution of the iterative training operation, the damage recognition graph neural network model is obtained. The iterative training operation includes: Select target training sample data from the training dataset; The graph structure training data in the target training sample data is input into the initial damage recognition graph neural network model to obtain the classification prediction results of the damage location and the regression prediction results of the damage severity. The cross-entropy loss is determined based on the classification prediction results and the damage location label data in the target training sample data, and the mean squared error loss is determined based on the regression prediction results and the damage severity label data in the target training sample data; the parameters of the initial damage recognition graph neural network model are updated based on the cross-entropy loss and the mean squared error loss.

8. A wind turbine blade damage identification device, characterized in that, include: The data acquisition module is used to acquire the raw strain signal data of the wind turbine blades; The data preprocessing module is used to perform adaptive mode decomposition, denoising, and reconstruction processing on the original strain signal data to obtain the reconstructed signal; The graph structure data module is used to divide the reconstructed signal into multiple signal segments and treat each signal segment as a graph node. Based on the signal features corresponding to the graph nodes, the similarity between each graph node is determined. Based on the similarity, the neighboring nodes of each graph node are determined. Based on the similarity, the edges between each graph node and its neighboring nodes are constructed. Based on each graph node and the edges, graph structure data is obtained. The damage identification module is used to determine the damage identification result of the wind turbine blade corresponding to the graph structure data based on a trained damage identification graph neural network model; wherein, the damage identification result includes the damage location and / or damage severity.

9. An electronic device, characterized in that, The method includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the wind turbine blade damage identification method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that, when executed by a processor, implement the wind turbine blade damage identification method as described in any one of claims 1 to 7.

Images