CN-120635057-B - Method and system for identifying damage of blade of miniature aero-engine

CN120635057BCN 120635057 BCN120635057 BCN 120635057BCN-120635057-B

Abstract

The invention discloses a method and a system for identifying damage of a blade of a miniature aeroengine, wherein the method comprises the steps of S1, generating an initial data set by using multi-source data of the surface and the internal structure of the blade of the miniature aeroengine, S2, preprocessing the initial data set to obtain a preprocessed data set, S3, training an improved YOLOv S algorithm by using the preprocessed data set, optimizing the trained improved YOLOv S algorithm to obtain a trained improved YOLOv S algorithm, and S4, identifying and monitoring the multi-source data of the surface and the internal structure of the blade of the miniature aeroengine, which are acquired in real time, by using the trained improved YOLOv S algorithm to complete identification of the blade damage of the miniature aeroengine.

Inventors

Huo Xianglong
WANG LIN
WANG DONG

Assignees

保定市玄云涡喷动力设备研发有限公司

Dates

Publication Date: 20260512
Application Date: 20250716

Claims (5)

1. A method for identifying damage to a blade of a miniature aeroengine, the method comprising: s1, generating an initial data set by using multi-source data existing on the surface and the internal structure of a blade of a microminiature aeroengine; s2, preprocessing the initial data set to obtain a preprocessed data set; step S3, training the improved YOLOv S algorithm by using the preprocessing data set, and optimizing the trained improved YOLOv S algorithm to obtain a trained improved YOLOv S algorithm; s4, performing identification monitoring on multisource data of the surface and internal structure of the miniature aero-engine blade acquired in real time by using a trained improved YOLOv S algorithm, and completing identification of damage of the miniature aero-engine blade; the improved YOLOv s algorithm specifically comprises the following steps: The improved YOLOv s algorithm comprises an input end, a main network, a neck and a head; the input is used for receiving the preprocessing data set; Based on YOLOv s algorithm, the backbone network uses a ConvNeXt V2 feature extraction module to replace 2C 3 modules at the tail end of the backbone network of YOLOv s algorithm, and a person SimAM is embedded at the rear end of each C3 module of the backbone network without a attention mechanism; The neck adopts an FPN structure and PANet networks, the FPN structure provides multi-scale feature expression, and the PANet networks aggregate information paths among different feature layers, so that effective fusion of features is realized, and the integrity and diversity of the features are maintained; The head adopts 3 convolution layers to carry out convolution prediction; the pretreatment method in the step S2 specifically comprises the following steps: S21, filtering and denoising the initial data set to obtain a first processed image; According to the preliminary multi-source data set, denoising and edge enhancement processing are carried out on image data of the appearance form of the blade by adopting an image processing technology, a clear surface characteristic image is formed, and whether obvious damage marks exist on the surface of the blade is determined; if damage marks are detected in the surface feature images, carrying out frequency spectrum decomposition on dynamic change information through a vibration signal analysis method to obtain abnormal vibration modes of the internal structure of the blade, and judging whether damage hidden danger caused by internal change exists or not; the method comprises the steps of comprehensively comparing a surface characteristic image with an abnormal vibration mode, and classifying and processing damage initial data by adopting a support vector machine algorithm to obtain a preliminary division result of damage types; according to the preliminary division result of the damage types, extracting characteristics of different types of damage data, acquiring specific positions and morphological characteristics of the damage parts, and determining the distribution range of the damage; s22, dividing the first processed image by using an image division model to obtain a second processed image containing damage information; And S23, performing damage identification on the second processed image by using the improved ResNet network to obtain a third processed image containing damage identification information, namely a preprocessed data set.
2. The method according to claim 1, wherein in the step S22, the image segmentation model is as follows: extracting features of the first processed image; polymerizing the extracted feature map to obtain polymerized features; And processing key information in the aggregation characteristics by adopting a full connection method and a non-maximum suppression technology, identifying damage information and dividing the damage information.
3. A micro-aeroengine blade damage identification system for implementing the damage identification method according to any one of claims 1-2, characterized in that the system comprises: the data acquisition module is used for generating an initial data set by using multi-source data existing on the surface and the internal structure of the blade of the miniature aero-engine; The data preprocessing module is used for preprocessing the initial data set to obtain a preprocessed data set; the model optimization module is used for training the improved YOLOv s algorithm by using the preprocessing data set and optimizing the trained improved YOLOv s algorithm to obtain a trained improved YOLOv s algorithm; The damage identification module is used for identifying and monitoring multisource data of the surface and the internal structure of the blade of the miniature aero-engine acquired in real time by using a trained improved YOLOv s algorithm, so as to complete the damage identification of the blade of the miniature aero-engine; the improved YOLOv s algorithm specifically comprises the following steps: The improved YOLOv s algorithm comprises an input end, a main network, a neck and a head; the input is used for receiving the preprocessing data set; Based on YOLOv s algorithm, the backbone network uses a ConvNeXt V2 feature extraction module to replace 2C 3 modules at the tail end of the backbone network of YOLOv s algorithm, and a person SimAM is embedded at the rear end of each C3 module of the backbone network without a attention mechanism; The neck adopts an FPN structure and PANet networks, the FPN structure provides multi-scale feature expression, and the PANet networks aggregate information paths among different feature layers, so that effective fusion of features is realized, and the integrity and diversity of the features are maintained; the header performs convolution prediction using 3 convolution layers.
4. The micro-miniature aeroengine blade damage identification system according to claim 3, wherein in the data preprocessing module, the preprocessing process specifically comprises: filtering and denoising the initial data set to obtain a first processed image; Dividing the first processed image by using an image division model to obtain a second processed image containing damage information; and performing damage identification on the second processed image by using the improved ResNet network to obtain a third processed image containing damage identification information, namely a preprocessed data set.
5. The micro-miniature aeroengine blade damage identification system of claim 4, wherein the image segmentation model is: extracting features of the first processed image; polymerizing the extracted feature map to obtain polymerized features; And processing key information in the aggregation characteristics by adopting a full connection method and a non-maximum suppression technology, identifying damage information and dividing the damage information.

Description

Method and system for identifying damage of blade of miniature aero-engine Technical Field The invention relates to the technical field of damage detection, in particular to a method and a system for identifying damage to a blade of a miniature aeroengine. Background The field of research of micro-aeroengines plays a vital role in the aeronautical industry, the health of its core components, such as blades, being directly related to the safety of flight and to the life of the equipment. With the rapid development of aviation technology, ensuring the reliability of these critical components in complex working environments has become a focus of industry attention. The blade is used as a part which bears high load and high stress in the engine and is extremely easy to be influenced by various damages, so that the accurate identification of the damages is not only a requirement for technical progress, but also a foundation stone for guaranteeing safe operation. However, current methods for blade damage identification still have significant drawbacks. Many traditional approaches often rely on a single detection mode or a fixed analysis model, and are difficult to adapt to the diversity of different types of injuries in morphology and performance, and especially when facing small changes in complex environments, misjudgment or missed judgment is easy to occur. In addition, the existing method often has a great deal of trouble in processing multi-source data fusion and dynamic feature extraction, so that deep features of damage are not fully mined, and the recognition precision and efficiency are limited. In this context, the core challenges faced in this field are increasingly manifest. The first problem is how to extract representative features from the damage to the blade, which features need to cover not only the appearance of the damage, but also the intrinsic law of variation, which process often becomes exceptionally difficult due to the diversity and complexity of the type of damage. Further, due to the lack of systematic generalization and digital expression of different lesion characteristics, it is difficult to form a unified standard to rapidly distinguish and match lesion types, which directly affects the accuracy and real-time of recognition. Therefore, how to construct a method capable of comprehensively extracting multi-dimensional characteristics of blade damage and generating unique digital identifications so as to realize rapid and accurate matching of different damage types becomes a key problem to be solved in the current research. Disclosure of Invention In order to solve the technical problems, the invention provides a method and a system for identifying damage of a blade of a miniature aero-engine, wherein the method specifically comprises the following steps: s1, generating an initial data set by using multi-source data existing on the surface and the internal structure of a blade of a microminiature aeroengine; s2, preprocessing the initial data set to obtain a preprocessed data set; step S3, training the improved YOLOv S algorithm by using the preprocessing data set, and optimizing the trained improved YOLOv S algorithm to obtain a trained improved YOLOv S algorithm; and S4, performing identification and monitoring on multisource data of the surface and internal structure of the miniature aero-engine blade acquired in real time by using a trained and improved YOLOv S algorithm, and completing identification of damage of the miniature aero-engine blade. Optionally, the preprocessing method in step S2 specifically includes: S21, filtering and denoising the initial data set to obtain a first processed image; s22, dividing the first processed image by using an image division model to obtain a second processed image containing damage information; And S23, performing damage identification on the second processed image by using the improved ResNet network to obtain a third processed image containing damage identification information, namely a preprocessed data set. Optionally, in the step S23, the image segmentation model is: extracting features of the first processed image; polymerizing the extracted feature map to obtain polymerized features; And processing key information in the aggregation characteristics by adopting a full connection method and a non-maximum suppression technology, identifying damage information and dividing the damage information. Optionally, in the step S3, the improved YOLOv S algorithm specifically includes: The improved YOLOv s algorithm comprises an input end, a main network, a neck and a head; the input is used for receiving the preprocessing data set; the backbone network uses a ConvNeXt V2 feature extraction module to replace 2C 3 modules at the tail end of the backbone network of the YOLOv5s algorithm on the basis of YOLOv s algorithm; The neck is characterized by adopting an FPN structure and PANet networks; the header performs convolution prediction using 3 con