CN-121686077-B - Wind blade damage defect detection method and system based on vision

CN121686077BCN 121686077 BCN121686077 BCN 121686077BCN-121686077-B

Abstract

The invention provides a method and a system for detecting damage defects of wind blades based on vision, and relates to the field of image recognition, wherein the method comprises the steps of constructing a damage association knowledge graph based on working environment characteristics and damage defects of a plurality of sample wind blades; the improved YOLOv detection network is constructed and trained, wherein the improved YOLOv detection network comprises an adaptive input unit, the adaptive input unit is used for adjusting the resolution of an image based on the image characteristics and damage association knowledge graph of the image, constructing damage detection paths corresponding to various working environment types based on the damage association knowledge graph and the improved YOLOv detection network, determining the damage detection paths based on the working environment characteristics of the wind blade to be detected and the damage detection paths corresponding to the various working environment types, and detecting damage defects of the wind blade to be detected based on the damage detection paths and the improved YOLOv detection network.

Inventors

LV CAIXIA
WU FEIFEI
SU JIE
GUO XIAODONG
CHANG HONG

Assignees

内蒙古农业大学职业技术学院

Dates

Publication Date: 20260512
Application Date: 20251212

Claims (10)

1. The method for detecting the damage defect of the wind blade based on vision is characterized by comprising the following steps of: Acquiring working environment characteristics and damage defects of a plurality of sample wind blades, wherein the working environments of the plurality of sample wind blades are different; Constructing a damage association knowledge graph based on the working environment characteristics and damage defects of the wind power blades of the plurality of samples, wherein the damage association knowledge graph is used for recording damage association relations of a plurality of parts of the wind power blades corresponding to a plurality of working environment types; Constructing and training an improved YOLOv detection network, wherein the improved YOLOv detection network comprises an adaptive input unit, and the adaptive input unit is used for adjusting the resolution of an image based on the image characteristics and the damage association knowledge graph of the image; Constructing damage detection paths corresponding to various working environment types based on the damage association knowledge graph and the improved YOLOv detection network, wherein the damage detection paths are used for determining the detection sequence of a plurality of parts of the wind blade; acquiring an image and working environment information of a wind blade to be detected; Determining a damage detection path based on the working environment characteristics of the wind blade to be detected and damage detection paths corresponding to various working environment types; and detecting the damage defect of the wind blade to be detected based on the damage detection path and the improved YOLOv detection network.
2. The vision-based wind blade damage defect detection method of claim 1, wherein constructing the damage-related knowledge graph based on the operational environmental characteristics and damage defects of the plurality of sample wind blades comprises: clustering the plurality of sample wind power blades based on the working environment characteristics of the plurality of sample wind power blades to determine a plurality of blade classes, wherein one blade class corresponds to one working environment type; For each blade class, determining damage association coefficients of any two parts of the wind blade based on damage defects of a plurality of sample wind blades included in the blade class, determining damage association parts of each part corresponding to the blade class based on the damage association coefficients of any two parts of the wind blade corresponding to each blade class, and constructing a sub-graph corresponding to the blade class; And constructing a damage association knowledge graph based on the sub-graph corresponding to each leaf class.
3. The vision-based wind blade damage defect detection method of claim 2, wherein the adaptive input unit adjusts the resolution of the image based on the image features and damage-associated knowledge-graph of the image, comprising: Performing target positioning on an image, and extracting image features of the image, wherein the image features at least comprise background complexity, target density and average target size; determining scene complexity of the image based on image features of the image; determining a component corresponding to the image; Obtaining a damage defect detection result of a damage-related part of the part corresponding to the image; and adjusting the resolution of the image based on the scene complexity of the image and the damage defect detection result of the damage-associated component of the component corresponding to the image.
4. The vision-based wind blade damage defect detection method of claim 3, wherein constructing damage detection paths corresponding to a plurality of working environment types based on the damage correlation knowledge graph and the improved YOLOv detection network comprises: for each working environment type, determining sampling probability of each component based on a sub-map corresponding to the blade class and damage defects of a plurality of sample wind blades included in the blade class; Carrying out population initialization based on the sampling probability and damage association knowledge graph of each component, wherein the population comprises a plurality of individuals, and each individual corresponds to a damage detection path; Constructing an adaptability function, wherein the adaptability function is related to the damage detection efficiency and the pixel point sum corresponding to the individual; Sampling from a plurality of sample wind blades included in the blade class, obtaining a plurality of sampled sample wind blades; for each individual, generating damage detection results of each sampled sample wind blade based on the damage detection path corresponding to the individual through the improved YOLOv detection network; Calculating an fitness value of each individual based on the damage detection result and the fitness function of each sampled sample wind blade; And (3) based on the fitness value of each individual, selecting, crossing and mutating, updating the population, and performing iterative optimization until the termination condition is met.
5. The vision-based wind blade damage defect detection method of claim 4, wherein determining the sampling probability for each component based on the sub-map corresponding to the blade class and the damage defects of the plurality of sample wind blades included in the blade class comprises: Determining a damage defect probability for each component based on damage defects of a plurality of sample wind blades comprised by the blade class; and determining sampling probability of each part based on the damage defect probability of the part and the damage defect probability of the damage-related part of the part based on the sub-graph corresponding to the blade class.
6. The vision-based wind blade damage defect detection method of claim 5, wherein population initialization based on the sampling probability and damage-related knowledge-graph of each component comprises: Constructing a path constraint condition, wherein the path constraint condition at least comprises a damage-related component, a component repetition constraint and a component coverage constraint of two adjacent components in a damage detection path; for each individual, a current probability for each component is generated, and the individual is generated based on the current probability and sampling probability for each component and the path constraints.
7. The method for detecting damage defects of wind blades based on vision as recited in any one of claims 1 to 6, wherein the middle layer of the improved YOLOv detection network includes an initial feature fusion unit, an iterative feature fusion unit, a multi-scale interaction unit and a feature refining and outputting unit, wherein the initial feature fusion unit is used for performing scale expansion on a multi-scale feature map output by a backbone network of the improved YOLOv detection network to generate an expanded feature map, the iterative feature fusion unit is used for generating three sets of feature maps after bidirectional fusion based on the expanded feature map, the multi-scale interaction unit is used for performing cross-scale feature interaction and symmetric scale transformation on the three sets of feature maps after bidirectional fusion to generate a processed multi-scale feature map, and the feature refining and outputting unit is used for performing multi-layer refining and scale expansion on the processed multi-scale feature map to generate five sets of feature maps.
8. The method for detecting damage defects of wind blades based on vision as recited in any one of claims 1 to 6, wherein the detection head of the improved YOLOv detection network comprises a feature preprocessing unit, a spatial adaptive up-sampling unit, a dual-path feature fusion unit, a four-stage progressive feature fusion network and a three-stage optimized feature map output layer, wherein the feature preprocessing unit is used for performing dynamic receptive field adjustment and feature enhancement on five groups of feature maps output by an intermediate layer of the improved YOLOv detection network to generate a semantically unified basic feature map, the spatial adaptive up-sampling unit is used for up-sampling the basic feature map to generate an up-sampling feature map, the dual-path feature fusion unit is used for fusing the basic feature map and the up-sampling feature map to generate a semantic focusing feature map, the four-stage progressive feature fusion network is used for generating a multi-stage enhanced feature map based on the semantic focusing feature map, and the three-stage optimized feature map output layer is used for generating the fused feature map based on the multi-stage enhanced feature map.
9. The vision-based wind blade damage defect detection method of claim 5, wherein the loss function for training the modified YOLOv detection network includes damage classification loss and bounding box loss, wherein the damage classification loss is related to sampling probability for each component corresponding to each working environment type.
10. A vision-based wind blade damage defect detection system, wherein the vision-based wind blade damage defect detection method of claim 1 is applied, comprising: the data acquisition module is used for acquiring the working environment characteristics and damage defects of the plurality of sample wind blades, wherein the working environments of the plurality of sample wind blades are different; The map construction module is used for constructing a damage association knowledge map based on the working environment characteristics and damage defects of the plurality of sample wind blades, wherein the damage association knowledge map is used for recording damage association relations of a plurality of parts of the wind blades corresponding to a plurality of working environment types; The model building module is used for building and training a modified YOLOv detection network, wherein the modified YOLOv detection network comprises an adaptive input unit, and the adaptive input unit is used for adjusting the resolution of an image based on the image characteristics and the damage association knowledge graph of the image; The path optimization module is used for constructing damage detection paths corresponding to various working environment types based on the damage association knowledge graph and the improved YOLOv detection network, wherein the damage detection paths are used for determining the detection sequence of a plurality of parts of the wind blade; the data acquisition module is also used for acquiring images and working environment information of the wind turbine blade to be detected; The defect detection module is used for determining a damage detection path based on the working environment characteristics of the wind blade to be detected and damage detection paths corresponding to various working environment types; the defect detection module is also used for detecting the damage defect of the wind blade to be detected based on the damage detection path and the improved YOLOv detection network.

Description

Wind blade damage defect detection method and system based on vision Technical Field The invention relates to the field of image recognition, in particular to a method and a system for detecting damage defects of wind blades based on vision. Background The fan blade is used as a core functional component for capturing wind energy of the wind generating set, and the structural reliability of the fan blade directly determines the generating efficiency and the service cycle of the set. In an ideal state, the blades can efficiently convert wind energy into mechanical energy, and then drive the generator to generate electric energy. However, in a practical operating environment, fan blades are subject to many complex and harsh conditions. In extremely severe environments such as sand storm, a large amount of sand particles moving at high speed can continuously wash and impact the surface of the blade, so that the material on the surface of the blade is gradually worn out, and scratches, pits and other damages appear. Salt spray erosion is also a common corrosion condition, especially in coastal areas or offshore wind farms, salt in the air can form salt spray deposition on the surfaces of the blades, so that electrochemical corrosion is initiated, the protective coating on the surfaces of the blades is damaged, and the blade material gradually loses the original performance. In addition, the continuous alternating changes of temperature and humidity can also have adverse effects on the blade, and the difference of thermal expansion coefficients among different materials can lead to stress in the blade, so that microcracks can be generated and expanded under the long-term action. Under the combined action of the complex environmental stresses, progressive damage such as layering, cracks and the like is very easy to occur on the surface of the blade. These initial stages of injury may not be apparent, but progress over time to worsen. Once such a defect occurs, aerodynamic performance of the blade will be significantly reduced, resulting in a reduction in aerodynamic efficiency and a reduction in power generation. More seriously, the damage can also cause catastrophic structural damage, such as blade breakage, and the like, which not only causes huge economic loss, but also can form serious threat to the surrounding environment and personnel safety. At present, the operation and maintenance of the fan blade mainly adopts a traditional mode mainly comprising manual visual inspection. Although this approach can to some extent detect significant damage to the blade surface, there are a number of drawbacks that are difficult to overcome. On the one hand, the detection period of the manual visual inspection is too long. Because wind farms are generally widely distributed and have a large number of blades and high blade mounting positions, manual inspection requires a lot of time and labor costs. Moreover, to ensure the comprehensiveness of the inspection, it is often necessary to inspect all the blades one by one at regular intervals, which further lengthens the inspection period, making it difficult to find the damage of the blades in time during the interval between the two inspections. On the other hand, it is difficult to perform quantitative evaluation by manual visual inspection. The inspector mainly relies on experience to judge the damage degree of the blade, and lacks objective and accurate quantitative indexes. Different inspectors may have differences in the evaluation results of the same damage, which makes subsequent maintenance decisions difficult, and the maintenance priority and maintenance scheme cannot be accurately determined. Further, the manual visual inspection is insufficient for coverage of the high-risk area. Some blade parts, such as blade tips, blade roots and the like, are difficult to observe in a short distance by an inspector due to special positions, and have an inspection blind area. The damage in these high risk areas may be more severe, but it is likely to have more serious consequences due to the inability to be found in time. Therefore, it is desirable to provide a method and a system for detecting damage defects of wind blades based on vision, so as to realize automatic detection of damage of wind blades. Disclosure of Invention The invention provides a vision-based wind blade damage defect detection method, which comprises the steps of obtaining working environment characteristics and damage defects of a plurality of sample wind blades, constructing a damage association knowledge graph based on the working environment characteristics and the damage defects of the plurality of sample wind blades, wherein the damage association knowledge graph is used for recording damage association relations of a plurality of parts of the wind blades corresponding to a plurality of working environment types, constructing and training an improved YOLOv detection network, wherein the improved YOLOv