CN-121997071-A - Fan blade damage identification method, device, equipment and medium

CN121997071ACN 121997071 ACN121997071 ACN 121997071ACN-121997071-A

Abstract

The invention provides a fan blade damage identification method, device, equipment and medium, which relate to the technical field of wind power fault diagnosis and are used for obtaining a reconstruction signal by carrying out self-adaptive modal decomposition, denoising and reconstruction processing on original strain signal data, dividing the reconstruction signal into a plurality of signal segments, taking each signal segment as a graph node, determining similarity between each graph node according to signal characteristics corresponding to the graph nodes, determining neighbor nodes of each graph node based on the similarity, constructing edges between each graph node and the neighbor nodes of the graph node based on the similarity, obtaining graph structure data based on each graph node and the edges, and determining damage positions and/or damage severity of fan blades corresponding to the graph structure data based on a trained damage identification graph neural network model so as to realize single damage and multi-damage position positioning and severity assessment while improving the accuracy of damage identification.

Inventors

YI RAN
WANG QINGTIAN
ZHANG YONG
CHEN LIYU
ZHOU SIYU
TONG TONG
LI ZELI
YANG ZIYANG
HU JIE
ZHANG HUIJUN
GAO YI
SHEN YONGHONG

Assignees

华能重庆奉节风电有限责任公司
中国华能集团清洁能源技术研究院有限公司

Dates

Publication Date: 20260508
Application Date: 20260127

Claims (10)

1. The method for identifying damage of the fan blade is characterized by comprising the following steps of: Acquiring original strain signal data of a fan blade; performing adaptive modal decomposition, denoising and reconstruction processing on the original strain signal data to obtain a reconstructed signal; Dividing the reconstruction 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 characteristics corresponding to the graph nodes, determining the neighbor nodes of each graph node based on the similarity, constructing the edges between each graph node and the neighbor nodes based on the similarity, and obtaining graph structure data based on each graph node and the edges; and determining a damage identification result of the fan blade corresponding to the graph structure data based on the trained damage identification graph neural network model, wherein the damage identification result comprises a damage position and/or damage severity.
2. The method for identifying damage to a fan blade according to claim 1, wherein performing adaptive modal decomposition, denoising and reconstruction processing on the original strain signal data to obtain a reconstructed signal comprises: Performing self-adaptive optimization on the parameters of the successive variation modal decomposition by adopting a meta heuristic optimization algorithm to obtain optimized parameters; Decomposing the original strain signal data by adopting successive variation modal decomposition based on the optimized parameters to obtain a plurality of eigenvalue function components; screening and denoising the plurality of eigenvalue function components to obtain denoised component signals; and carrying out superposition processing on the denoised component signals to obtain the reconstruction signals.
3. The method for identifying damage to fan blades according to claim 2, wherein the adaptive optimization of the parameters of the successive variation modal decomposition by using a meta-heuristic optimization algorithm, the obtaining of the optimized parameters comprises: And performing self-adaptive optimization on the penalty factor parameters of the successive variation modal decomposition by adopting a sea image optimization algorithm to obtain optimized penalty factor parameters, wherein the optimized penalty factor parameters are obtained by minimizing the envelope entropy of the eigenvalue function components after decomposition.
4. The method of claim 1, wherein dividing the reconstructed signal into a plurality of signal segments comprises: And dividing the reconstructed signal by adopting a non-overlapping sliding window to obtain a plurality of signal fragments.
5. The method of claim 4, wherein said determining a neighbor node for each of said graph nodes based on said similarity comprises: and determining a neighbor node set by adopting a K neighbor algorithm based on the similarity between the graph node and the rest graph nodes.
6. The method of claim 5, wherein determining edges between each graph node and its neighbor nodes based on the similarity comprises: based on each neighbor node in the neighbor node set, determining edge weights between the graph node and the neighbor nodes by adopting a Gaussian kernel function according to the similarity between the graph node and the neighbor nodes; Based on the edge weights, edges between the graph nodes and the neighbor nodes are determined.
7. The method for identifying damage to fan blades according to claim 1, characterized by further comprising: The method comprises the steps of obtaining a training data set, wherein the training data set comprises a plurality of training sample data, each training sample data comprises graph structure training data and damage position label data and damage severity label data corresponding to the graph structure training data, the graph structure training data is obtained by carrying out self-adaptive modal decomposition, denoising and reconstruction processing on original strain signal data, dividing the reconstruction signal into a plurality of signal segments, taking each signal segment as a graph node, determining similarity between each graph node according to signal characteristics corresponding to the graph node, determining neighbor nodes of each graph node based on the similarity, constructing edges between each graph node and the neighbor nodes, and obtaining edges based on each graph node and the edges; Performing iterative training operation on the initial damage recognition graph neural network model based on the training data set until the iterative training termination condition is determined to be met, and obtaining the damage recognition graph neural network model based on parameters of the initial damage recognition graph neural network model updated when the iterative training operation is performed last time, wherein the iterative training operation comprises: selecting target training sample data from the training data set; inputting the graph structure training data in the target training sample data into the initial damage identification graph neural network model to obtain a classification prediction result of a damage position and a regression prediction result of damage severity; Determining cross entropy loss based on the classification prediction result and damage position label data in the target training sample data, determining mean square error loss based on the regression prediction result and damage severity label data in the target training sample data, and updating parameters of the initial damage identification graph neural network model based on the cross entropy loss and the mean square error loss.
8. A fan blade damage identification device, comprising: the data acquisition module is used for acquiring original strain signal data of the fan blade; the data preprocessing module is used for carrying out self-adaptive modal decomposition, denoising and reconstruction processing on the original strain signal data to obtain a reconstruction signal; The graph structure data module is used for dividing the reconstruction 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 characteristics corresponding to the graph nodes, determining the neighbor nodes of each graph node based on the similarity, constructing edges between each graph node and the neighbor nodes based on the similarity, and obtaining graph structure data based on each graph node and the edges; The damage identification module is used for determining damage identification results of the fan blades corresponding to the graph structure data based on the trained damage identification graph neural network model, wherein the damage identification results comprise damage positions and/or damage severity.
9. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the fan blade damage identification method of any of claims 1 to 7 when the computer program is executed.
10. A computer readable storage medium storing computer instructions which, when executed by a processor, implement the method of identifying fan blade damage according to any one of claims 1 to 7.

Description

Fan blade damage identification method, device, equipment and medium Technical Field The invention relates to the technical field of wind power fault diagnosis, in particular to a method, a device, equipment and a medium for identifying damage of a fan blade. Background The existing offshore wind turbine blade damage identification method is mainly divided into a machine vision type method and a response signal type method, wherein in terms of signal processing and damage identification algorithm, the traditional machine learning algorithm (such as a support vector machine and a K nearest neighbor method) has the problems of low precision and poor generalization capability, the existing deep learning algorithm (such as a neural network and a long-term and short-term memory network) has the problems of large damage positioning and severity identification errors caused by weak modeling capability when blade damage data of a non-Euclidean structure are processed, and damage identification accuracy is greatly reduced when the damage positioning and severity identification errors face strong noise interference in a marine environment. Disclosure of Invention In view of the above, the present invention aims to provide a method, a device and a medium for identifying damage to a fan blade, so as to improve the accuracy of damage identification and simultaneously realize the position location and severity assessment of single damage and multiple damage. In a first aspect, the present application provides a method for identifying damage to a fan blade, including: Acquiring original strain signal data of a fan blade; performing adaptive modal decomposition, denoising and reconstruction processing on the original strain signal data to obtain a reconstruction signal; Dividing a 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 characteristics corresponding to the graph nodes, determining the neighbor nodes of each graph node based on the similarity, constructing edges between each graph node and the neighbor nodes based on the similarity, and obtaining graph structure data based on each graph node and the edges; And determining a damage identification result of the fan blade corresponding to the graph structure data based on the trained damage identification graph neural network model, wherein the damage identification result comprises a damage position and/or damage severity. Optionally, performing adaptive modal decomposition, denoising and reconstruction processing on the original strain signal data to obtain a reconstructed signal, including: Performing self-adaptive optimization on the parameters of the successive variation modal decomposition by adopting a meta heuristic optimization algorithm to obtain optimized parameters; Decomposing the original strain signal data by adopting successive variation modal decomposition based on the optimized parameters to obtain a plurality of eigenmode function components; screening and denoising the plurality of eigenmode function components to obtain denoised component signals; and (3) carrying out superposition processing on the denoised component signals to obtain a reconstructed signal. Optionally, a meta heuristic optimization algorithm is adopted to adaptively optimize the parameters of the successive variation modal decomposition, and the optimized parameters comprise: And performing self-adaptive optimization on the penalty factor parameters of the successive variation modal decomposition by adopting a sea image optimization algorithm to obtain optimized penalty factor parameters, wherein the optimized penalty factor parameters are the minimum envelope entropy of the decomposed intrinsic modal function components. Optionally, dividing the reconstructed signal into a plurality of signal segments comprises: And dividing the reconstructed signal by adopting a non-overlapping sliding window to obtain a plurality of signal fragments. Optionally, determining, for each graph node, its neighboring nodes based on the similarity, including: Based on the similarity between the graph node and the rest of the graph nodes, a K neighbor algorithm is adopted to determine a neighbor node set. Optionally, determining an edge between each graph node and its neighbor node based on the similarity includes: Based on each neighbor node in the neighbor node set, determining the edge weight between the graph node and the neighbor node by adopting a Gaussian kernel function according to the similarity between the graph node and the neighbor node; Based on the edge weights, edges between the graph nodes and the neighbor nodes are determined. Optionally, the method for identifying damage to a fan blade provided by the application further comprises the following steps: The method comprises the steps of obtaining a training data set, wherein the training data set comprises a