CN-121351497-B - Intelligent characterization and service life assessment method for surface defects of aero-engine structure

CN121351497BCN 121351497 BCN121351497 BCN 121351497BCN-121351497-B

Abstract

The invention discloses an intelligent characterization and service life assessment method for surface defects of an aero-engine structure, and relates to the technical field of aero-engine structure health management. The method comprises the steps of collecting defect data through a multi-source sensing technology in a fusion mode, constructing a knowledge graph, utilizing a multi-branch deep learning model to achieve automatic identification of defect types and intelligent extraction of key geometric features, determining standard defect sizes based on a statistical distribution model of group defect data, rapidly calculating defect stress concentration coefficients through a trained AI proxy model to achieve mechanical equivalence of irregular defects, and finally dynamically simulating and predicting evolution paths and residual lives of equivalent defect implantation component digital twin models. The invention solves the technical problems of dependence on experience of characterization, isolation of analysis links and static conservation evaluation, realizes the intelligent treatment from accurate digital characterization to individual life prediction of the surface defects of the engine, and provides support for the condition-dependent maintenance and predictive health management of the engine.

Inventors

HU XIAOAN
YU CHENGLONG
YANG QINZHENG
TENG XUEFENG
SUI TIANXIAO
LI JIAN
LIU FENCHENG
NIE XIANGFAN

Assignees

南昌航空大学

Dates

Publication Date: 20260512
Application Date: 20251030

Claims (6)

1. The intelligent characterization and life assessment method for the surface defects of the aero-engine structure is characterized by comprising the following steps of: acquiring multi-modal defect data of the surface of a target aero-engine; The method comprises the steps of establishing a target aeroengine surface defect knowledge graph by fusing context information of multi-modal defect data, wherein the context information of the multi-modal defect data comprises part models, specific positions, material brands, service histories and history maintenance records of defects; identifying multi-modal defect data based on the trained multi-branch deep convolutional neural network model to obtain geometrical characteristic parameters of surface defects of the aero-engine; Characterizing the surface defect of the target aeroengine as a standard equivalent crack identical to the mechanical effect through a trained AI proxy model based on the geometrical characteristic parameters of the surface defect of the aeroengine, wherein the AI proxy model is a rapid mapping model from the defect geometrical parameters to a theoretical stress concentration coefficient Kt established based on a lightweight neural network; Based on standard equivalent cracks and a knowledge graph, carrying out dynamic simulation by a digital twin model of the aero-engine component and combining an actual load spectrum of the aero-engine component, predicting an evolution path of the surface defect of the target aero-engine and evaluating the residual service life; The method for predicting the evolution path of the surface defect of the target aero-engine and evaluating the residual service life specifically comprises the following steps: based on SIMULIA Abaqus platforms, constructing a multi-physical field coupling model through coupling structure dynamics, fatigue damage mechanics and corrosion evolution physics; Implanting standard equivalent cracks into the multi-physical field coupling model, and calculating crack expansion quantity under each load cycle through a Paris formula or a Forman formula based on fatigue crack expansion theory of fracture mechanics; When the crack propagation reaches the critical crack size, the residual life of the crack is obtained.
2. The method for intelligently characterizing and evaluating the service life of the surface defects of the aero-engine structure according to claim 1, wherein the multi-branch deep convolutional neural network model specifically comprises: PointNet ++ branches for processing three-dimensional point cloud data; a CNN branch for processing two-dimensional image data; A connection layer for feature fusion; and an output layer for feature output.
3. The method for intelligently characterizing and evaluating the service life of the surface defect of the aero-engine structure according to claim 1, wherein the identifying the multi-modal defect data based on the trained multi-branch deep convolutional neural network model to obtain the geometrical characteristic parameters of the surface defect of the aero-engine comprises the following steps: Performing feature extraction and downsampling on the three-dimensional point cloud data through a Set extraction (SA) module of PointNet ++, and performing upsampling and feature fusion through a Feature Propagation (FP) module to obtain a global feature vector of the surface defect of the aeroengine; extracting feature vectors from the two-dimensional image through CNN branches; splicing the global feature vector output by PointNet ++ branch with the feature vector output by CNN branch to obtain a fusion feature vector of the surface defect of the aero-engine; The fusion feature vector obtains probability distribution of defect types through a full-connection layer with the output dimension being the defect category number and a Softmax function; And the fusion feature vector obtains the feature value of the specific aeroengine surface defect through the full-connection layer with the output dimension being the number of the required geometric features.
4. The intelligent characterization and life assessment method for surface defects of an aero-engine structure according to claim 1, wherein the aero-engine surface defects comprise scratch defects, pit defects and corrosion defects.
5. The method for intelligently characterizing and evaluating the service life of the surface defects of the aero-engine structure according to claim 1, wherein the rapid mapping model from the geometric parameters of the defects to the theoretical stress concentration coefficient Kt based on the lightweight neural network specifically comprises the following steps: calculating Kt values of defect models with different geometric forms in batches by a parameterized finite element method, and generating massive defect geometric-Kt value sample pairs; And training a lightweight neural network regression model by taking the sample pair as a training set, wherein the model learns a nonlinear mapping relation from the geometric parameters of the defects to the Kt values.
6. The intelligent characterization and life assessment method for the surface defects of the aero-engine structure according to claim 1, wherein when the intelligent characterization method for the defects is constructed, the method further comprises: obtaining depth data of a plurality of defects of the same type based on the knowledge graph; performing statistical analysis on the extracted depth data of the defects of the same type; And performing data fitting on the depth data of the defects of the same type by using the mixed probability model, and determining standard representation size intervals of the defects according to a preset coverage rate criterion.

Description

Intelligent characterization and service life assessment method for surface defects of aero-engine structure Technical Field The application relates to the technical field of aeroengine structure health management, in particular to an intelligent characterization and service life assessment method for surface defects of an aeroengine structure. Background In the process of manufacturing, assembling and servicing key components such as blades, discs and casings of aeroengines, the surfaces of the key components are easy to produce micro defects such as scratches, pits and corrosion due to factors such as tool collision, foreign matter impact and corrosion environment. These defects are potential sources of fatigue cracks, severely threatening the structural safety and service life of the engine. Thus, in structural integrity analysis, it must be accurately characterized and evaluated. In the current engineering practice, the defect characterization method is over-experienced and simplified, generally relies on manual detection and judgment, and uses regular geometric shapes (such as semi-cylinders and hemispheres) to approximate complex real defect morphology, the simplification ignores key local characteristics of defects, so that input conditions of subsequent mechanical analysis are distorted, evaluation results are often over-conservative or potential safety hazards exist, in addition, defect data analysis is in an 'island' state, links such as defect detection, measurement, classification and service life evaluation are mutually disjointed, massive detection data are not effectively integrated and mined, a standard defect spectrum with statistical significance for guiding design cannot be extracted, the safety evaluation process is static, the traditional defect tolerance analysis is based on a fixed initial defect size, and dynamic evolution process of defects under real and complex loads cannot be simulated, so that residual service life prediction for a single part and specific defects of the single part cannot be realized, and the maintenance decision lacks foresight. In summary, there is still a great room for improvement in the characterization of surface defects of the current aeroengine structure in terms of authenticity and accuracy. Disclosure of Invention Based on the above, it is necessary to provide an intelligent characterization and life assessment method for the surface defects of the aero-engine structure. The technical scheme adopted in the specification is as follows: The specification provides an aeroengine structure surface defect intelligent characterization and life assessment method, which comprises the following steps: the method comprises the steps of obtaining multi-modal defect data of the surface of a target aero-engine, and constructing a target aero-engine surface defect knowledge graph by fusing context information of the multi-modal defect data; identifying multi-modal defect data based on the trained multi-branch deep convolutional neural network model to obtain geometrical characteristic parameters of surface defects of the aero-engine; Characterizing the surface defect of the target aeroengine as a standard equivalent crack identical to the mechanical effect through a trained AI proxy model based on the geometrical characteristic parameters of the surface defect of the aeroengine, wherein the AI proxy model is a rapid mapping model from the defect geometrical parameters to a theoretical stress concentration coefficient Kt established based on a lightweight neural network; Based on the crack and the knowledge map of standard equivalent, dynamic simulation is carried out through a digital twin model of the aero-engine component and by combining with an actual load spectrum of the aero-engine component, the evolution path of the surface defect of the target aero-engine is predicted, and the residual service life is estimated. Further, the context information of the multi-mode defect data comprises the model number, specific position, material brand, service history and history maintenance record of the part where the defect is located. Further, the constructing the target aeroengine surface defect knowledge graph includes: and (3) associating the entity nodes of the surface defects of the target aero-engine with the entity nodes of the parts, the entity nodes of the materials and the entity nodes of the working conditions through the relation edges, and constructing an interconnection and intercommunication knowledge graph of the surface defects of the target aero-engine. Further, the multi-branch deep convolutional neural network model specifically includes: PointNet ++ branches for processing three-dimensional point cloud data; a CNN branch for processing two-dimensional image data; A connection layer for feature fusion; and an output layer for feature output. Further, identifying the multi-modal defect data based on the trained multi-branch deep convolutional neural n