Aircraft surface structure damage infrared feature simulation and sample generation method

Abstract

The invention belongs to the technical field of numerical simulation and damage intelligent detection, and relates to an aircraft surface structure damage infrared feature simulation and sample generation method. According to the method, through the design of the virtual interface layer and the secondary homogenization, the magnitude order improvement of the full-size transient thermal simulation efficiency is realized while the physical precision is ensured, and batch simulation of a large number of damaged working conditions becomes possible. The physical effect of a sensor of the infrared camera is simulated through high fidelity, so that a simulation image is highly consistent with real external field data in the aspects of noise, texture and other characteristics, and the'simulation-reality 'domain difference is effectively reduced. By directly associating the simulation process with the preset damage parameters, automatic, accurate and zero-error generation of the truth value information is realized, and the problems of high manual labeling cost and poor consistency are fundamentally solved. According to the method, an efficient and reliable data generation way is provided for training a high-precision and high-robustness external field intelligent detection model, and the method has a definite engineering practical value.

Description

Technical Field

[0001] This invention belongs to the field of numerical simulation and intelligent damage detection technology, and relates to a method for simulating infrared features of surface structural damage on aircraft and generating samples. Background Technology

[0002] Special-purpose aircraft surface structures suffer various types of damage under severe thermal cycling loads, including cracking, ablation, and debonding. This damage not only directly threatens flight safety but also makes efficient and reliable damage detection and condition assessment of the structure after each flight mission a necessary and demanding task. However, existing detection methods have limitations in efficiency and accuracy, resulting in long assessment cycles and severely restricting the rapid support capabilities of aircraft. To overcome this bottleneck, damage detection technology needs to evolve towards intelligent methods. Currently, the mainstream technical approach to achieving intelligent damage detection relies on deep learning-based machine vision methods. The performance of these methods is highly dependent on a large-scale, high-quality, and accurately labeled training sample library. However, acquiring infrared samples of damage to thermal protection structures faces severe challenges: First, real damage samples are scarce, and obtaining them through actual flights or real tests is extremely costly and difficult to systematically cover all damage types, sizes, depths, and combinations of conditions. Second, sample annotation is difficult; damage features in infrared images are weak and have blurred boundaries, requiring extremely high professional costs for precise annotation and introducing subjective errors. Third, data generalization is poor; real data from a single source cannot cover complex and variable external environments, including different initial temperature differences, illumination, and observation angles. Although data enhancement can be achieved through traditional image transformations such as rotation, cropping, and color adjustment, this method only changes the appearance of the image and cannot generate new, physically realistic damage features, thus contributing little to improving the model's ability to identify and generalize to real defects. At the simulation technology level, existing numerical simulation methods for composite materials mostly focus on the mechanistic study of specific damage modes or the verification of isolated cases. In existing technologies, even when using homogenization methods based on representative volume units for multi-scale simulation of composite materials, the multi-level models constructed to accurately characterize damage mechanisms remain complex and computationally expensive when dealing with heterogeneous materials with complex Z-axis reinforcement processes, such as aircraft surface structures, including needle punching and sewing. This makes it difficult for traditional simulations to support high-throughput, parameterized calculations for hundreds or thousands of damage conditions, thus preventing their direct use in building large-scale training sample libraries. Furthermore, conventional simulations only output idealized temperature field cloud maps, lacking modeling of key sensor effects such as noise, resolution limitations, and non-uniformity in the real infrared camera imaging chain. This results in significant "domain differences" between simulated and real-world acquired data, leading to a sharp decline in the performance of models trained directly from these data in practical field applications. In summary, the contradiction between the high cost and low coverage of real-world sample acquisition and the high computational consumption and low data fidelity of traditional simulation methods is becoming increasingly prominent, becoming a major bottleneck restricting the development of rapid and intelligent infrared detection technology for thermal protection structures based on artificial intelligence. Therefore, there is an urgent need to develop a new method that can efficiently and automatically generate large-scale, high-fidelity, and precisely labeled infrared damage simulation samples while ensuring physical authenticity, so as to provide reliable data support for the training of intelligent detection models. Summary of the Invention

[0003] The purpose of this invention is to address the prominent problems faced by existing deep learning-based intelligent infrared detection methods for aircraft surface structural damage, including a lack of high-quality training samples, difficulty in obtaining them, high computational costs in simulation, and inconsistencies between the data and real-world infrared image features. To address this issue, this invention combines multi-scale efficient simulation, high-fidelity infrared sensor effect simulation, and a fully automated truth-value association process, providing a solution for batch generation of high-quality simulated infrared samples based on physical mechanisms. Compared to methods relying on limited real data or general image enhancement, the simulation method of this invention, through a virtual interface layer and secondary homogenization design, achieves an order-of-magnitude improvement in computational efficiency while maintaining accuracy, making batch simulation of hundreds or thousands of damage conditions possible. This enables the systematic generation of diverse damage samples with clear physical meaning, features closer to reality, and automatic and accurate truth-value association. The standardized damage infrared feature sample library, as the direct output of this method, provides a high-quality training foundation with sufficient data, reliable features, and accurate annotations for training intelligent damage recognition models for complex external environments.

[0004] The technical solution of this invention is as follows: A method for simulating and generating infrared features of surface structural damage on aircraft. First, by fusing intrinsic material properties and manufacturing process parameters, a multi-scale physical simulation model from mesoscopic to macroscopic is constructed, and the concept of a virtual interface layer is introduced to achieve efficient parameterized simulation of damage. Second, while ensuring the authenticity of the physical mechanism, a secondary homogenization strategy is used to significantly reduce computational complexity, thereby enabling transient heat conduction simulation of full-size structures and obtaining the real temperature field response of damage under external cooling conditions. Then, through high-fidelity infrared sensor imaging effect simulation, the temperature field is converted into a simulated infrared image that is highly consistent with the real acquired data in terms of noise, texture, and degradation features. Finally, based on preset damage parameters, damage ground truth values ​​containing category labels and spatial representation information are automatically generated for the simulated infrared image, forming sample pairs, and a standardized damage infrared feature sample library is constructed in a structured manner.

[0005] The generation method is specifically as follows: Step 1: Obtaining Multi-Scale Physical Simulation Parameters: Based on the process documents and material performance test results, obtain the set of multi-scale physical simulation parameters for the target thermal protection structure; Step 2: Construction of the mesoscopic statistical representative volumetric unit model: At the mesoscopic scale, statistical representative volumetric unit models of the needle-punched reinforced fiber fabric layer and the needle-punched reinforced fiber web layer are constructed respectively to calculate the equivalent thermophysical parameters of the two. Step 3, Calculation of equivalent thermal properties: The equivalent thermal conductivity of the two statistical representative volume element models, needle-punched reinforced fiber cloth and needle-punched reinforced fiber web, constructed in Step 2, is obtained in the three principal directions of the material through finite element simulation. The finite element simulation is to perform independent steady-state thermal analysis in the three orthogonal directions of X, Y, and Z for each statistical representative volume element model. Step 4, Virtual Interface Layer Attribute Definition: Based on the mesoscopic equivalent thermophysical parameters obtained in Step 3, define a virtual interface layer for efficient simulation of interlayer damage; Step 5, Mesoscale intermediate unit construction and explicit modeling of sewing thread: Based on the mesoscale equivalent thermal property parameters obtained in Step 3 and the virtual interface layer defined in Step 4, a mesoscale intermediate unit model is constructed. By integrating mesoscale equivalent properties and sewing process features, a geometric and physical foundation is laid for subsequent overall macroscopic homogenization calculations. Step 6: Calculation of overall equivalent thermal properties of mesoscopic intermediate units: Perform steady-state thermal analysis on the mesoscopic intermediate units constructed in Step 5, calculate their overall equivalent thermal conductivity in the three main material directions, and use them to homogenize the complex units into a set of orthogonal anisotropic macroscopic material parameters, and provide input for subsequent macroscopic simulation; realize the mesoscopic secondary homogenization after the mesoscopic homogenization.

[0006] Step 7, Macroscopic Structural Model Construction and Damage Parametric Simulation: First, based on the overall equivalent thermophysical parameters obtained in Step 6, a full-size macroscopic finite element model of the target structure is established; then, by parametrically modifying material properties, various types of preset damage are efficiently simulated, providing a computational basis for subsequent transient thermal analysis.

[0007] Step 8, Transient heat conduction simulation and temperature field extraction: Based on the macroscopic structural model constructed in Step 7, a transient heat conduction simulation of the real external environment thermal load is performed to obtain the response of the structural surface under dynamic thermal environment and extract its time-series temperature field dataset for subsequent infrared image generation. Step 9, Infrared Image Generation and Sensor Effect Simulation: The time-series surface temperature field dataset extracted in Step 8 is converted into a high-fidelity simulated infrared image sequence. By simulating the imaging physical process of a real infrared camera, the simulated data is made consistent with the real field acquisition data in terms of features. Step 10: Generation of Standardized Damage Infrared Feature Sample Library: First, define a set of damage conditions containing different damage types, locations, and size parameters; based on the full-size macroscopic finite element model of the target structure constructed in Step 7, according to the set of damage conditions, iteratively call the parametric damage simulation function in Step 7. In each iteration, based on a specific set of damage condition parameters, modify the material properties of the corresponding region of the model through a programmed script to efficiently generate a macroscopic finite element model with a specific damage configuration; then, batch execute the transient heat conduction simulation described in Step 8 on the generated set of macroscopic finite element models to obtain the time-series temperature field dataset corresponding to each model; then, batch execute the infrared image generation and sensor effect simulation process described in Step 9 on the time-series temperature field dataset to uniformly convert it into a simulated infrared image sequence; finally, perform automated post-processing and integration on the simulated infrared image sequence to construct a standardized damage infrared feature sample library that can be directly used for machine learning model training.

[0008] The parameter set in step 1 includes basic material parameters and key process parameters. The basic material parameters include the thermal conductivity, specific heat capacity, and density of the resin matrix and quartz fiber, as well as the surface emissivity of the composite material in the infrared band, used to define the material physical properties of all scale models. The key process parameters specifically include fiber reinforcement geometric parameters and Z-axis reinforcement process parameters. The fiber reinforcement geometric parameters cover the areal density, single-layer thickness, weaving structure, warp and weft yarn density, yarn cross-sectional shape and size, and fiber volume fraction of the fiber fabric layer, as well as the areal density, single-layer thickness, chopped fiber length range, and fiber volume fraction of the fiber web layer. The Z-axis reinforcement process parameters cover the needle punching density and needle punching depth of the needle punching process, as well as the sewing thread material, diameter, sewing pattern, and sewing spacing of the sewing process.

[0009] The statistical representative volumetric unit model of the needle-punched reinforced fiber fabric layer in step 2 consists of a resin matrix, a periodic plain-weave quartz fiber yarn structure embedded in the resin matrix, and Z-direction quartz needle-punched fibers randomly penetrating between the resin matrix and the yarn. The statistical representative volumetric unit model of the needle-punched reinforced mesh in step 2 consists of a resin matrix, quartz short-cut fibers distributed in the resin matrix at random positions and orientations, and Z-direction quartz needle-punched fibers that are also randomly distributed throughout.

[0010] The statistical representative volume element model in step 2 uses tetrahedral elements for spatial discretization, and the material properties of each component are defined according to the basic material parameters of the parameter set obtained in step 1.

[0011] The thickness direction, i.e. the Z-axis dimension, of the statistical representative volume unit model of the needle-punched reinforced fiber cloth layer and the statistical representative volume unit model of the needle-punched reinforced fiber web layer in step 2 is determined according to the single-layer thickness parameters of the fiber cloth layer and the fiber web layer defined in step 1, so as to characterize a complete single layer of material.

[0012] In step 2, the representative volume element model of the needle-punched reinforced fiber fabric layer has an in-plane dimension L in the XY direction. c It needs to simultaneously meet the requirements of periodic characterization of the braided structure and statistical representativeness of the needle-punched fibers. The calculation formula is as follows:

[0013] Wherein, P is the basic periodic dimension of the braided structure, in millimeters; k is the periodic quantity coefficient, a positive integer not less than 1, and its typical value range is 2 to 4 to ensure the stability of the equivalent properties; ρ is the needle density in the key process parameters obtained in step 1, in the number of needles per square millimeter; N is the number of needled fibers that need to be included in the statistical representative volume unit model of the needled reinforced fiber fabric layer, and the value range of N is 3 to 10 fibers to ensure statistical representativeness.

[0014] The dimension L of the in-plane direction of the representative volume element model of the needle-punched reinforced fiber mesh layer in step 2 is described in step 2. b The main consideration is to meet the statistical representativeness requirements of needle-punched fibers, and the calculation formula is as follows:

[0015] Wherein, ρ is the needle density in the key process parameters obtained in step 1, in units of needles per square millimeter, and N' is the number of needled fibers that need to be included in the statistical representative volume unit model of the needled reinforced fiber web. To ensure statistical representativeness, the value of N' ranges from 3 to 10 fibers.

[0016] The statistical representative volume unit model of the needle-punched reinforced fiber web in step 2 is implemented using an equivalent modeling method for the chopped fibers inside. In the actual process, the small-diameter, randomly arranged quartz fiber filaments are equivalent to a set of cylindrical rods with a uniform diameter of d millimeters and random distribution of length and spatial orientation under the statistical scale of this statistical representative volume unit model. This model strictly maintains consistency with the fiber volume fraction defined in step 1. d is usually taken as 0.05-0.2 to solve the problem that it is difficult to directly and accurately model due to its complex real shape.

[0017] Step 3, the steady-state thermal analysis, is performed sequentially for each specific direction. First, the equivalent thermal conductivity in the model thickness direction, defined as the Z-direction, is calculated. Then, a constant heat flux density q and a constant temperature T are applied to the two opposing surfaces characterizing this direction. refTemperature boundary conditions, T ref Typically, 0℃ is used; the other four sides of the model are set as adiabatic boundary conditions, or periodic boundary conditions are applied to simulate the infinite periodic expansion of the material in the plane. After obtaining the steady-state temperature field, the average temperature T of the surface to which the heat flux boundary is applied is extracted. hot Calculate the average temperature difference ΔT = T between the two surfaces. hot T ref According to the one-dimensional Fourier law of heat conduction, this statistic represents the equivalent thermal conductivity k of the volume element model in the current direction, i.e., the Z-direction. z It can be calculated using the following formula:

[0018] Where H is the characteristic dimension of the model in the calculation direction; repeat the above process to calculate and obtain the equivalent thermal conductivity k in the X and Y directions of the statistical representative volume element model. x With k y Finally, the equivalent thermal conductivity in the three principal directions of the statistical representative volume element model of the needle-punched reinforced fiber fabric is output. , , The equivalent thermal conductivity in the three principal directions of the statistical representative volume element model of the needle-punched reinforced fiber web layer. , , The parameters constitute the key material properties for homogenizing microscopic heterostructures, providing input for subsequent heat transfer simulations at the mesoscopic and macroscopic scales.

[0019] The virtual interface layer in step 4 is a conceptual thin layer, and its geometric position is defined between the needle-punched reinforced fiber cloth layer and the needle-punched reinforced fiber mesh layer. Regarding material properties, when simulating a structurally intact state, the equivalent thermal conductivity of the virtual interface layer in the three principal material directions (X, Y, and Z) is obtained by taking the arithmetic mean of the equivalent thermophysical properties of its adjacent fiber cloth layer and fiber mesh layer in that direction. The specific calculation formula is as follows:

[0020]

[0021]

[0022] in, , , These are the equivalent thermal conductivity coefficients of the virtual interface layer in the three main material directions of X, Y, and Z, respectively.

[0023] The structure of the unit model in step 5 is formed by periodically stacking multiple basic ply units along the thickness direction, i.e., the Z-direction. Each basic ply unit strictly follows a fixed sequence of "fiber cloth layer - virtual interface layer - fiber web layer - virtual interface layer". The material thermophysical parameters of the fiber cloth layer and the fiber web layer are respectively assigned to the equivalent thermal conductivity of the needle-punched reinforced fiber cloth layer calculated in step 3. , , Equivalent thermal conductivity of needle-punched reinforced fiber mesh ply , , The material properties of the virtual interface layer are based on those calculated in step 4. , , To assign.

[0024] In step 5, the size of the unit model is determined according to the principle of representativeness: its in-plane (XY) dimension must be greater than the sewing spacing defined in step 1 to ensure that the periodicity of the sewing pattern can be fully represented; its thickness, i.e. the Z-axis dimension, is composed of M of the above basic ply units stacked together to reasonably represent the thickness direction characteristics of the overall ply structure, while avoiding excessive complexity of the model. Generally, M can be 4-8.

[0025] In step 5, the unit model is used to perform explicit three-dimensional geometric modeling of the sewing thread based on the sewing pattern and sewing spacing in the key process parameters obtained in step 1. The sewing thread is established as a cylinder that runs through the entire thickness of the combined unit. Its spatial position and path are determined by the sewing pattern and sewing spacing, and its material properties are set according to the quartz fiber properties defined in step 1.

[0026] Step 6, steady-state thermal analysis, involves performing independent simulations sequentially along the three principal material directions (X, Y, Z) of the medium-sized element model. First, the overall equivalent thermal conductivity in the Z direction is calculated: a constant heat flux density q' boundary condition and a constant reference temperature T' are applied to the two opposing surfaces of the model in the Z direction (thickness direction). ref Boundary conditions, T' ref Typically, 0℃ is used; the other four sides are set as adiabatic boundary conditions; after obtaining the steady-state temperature field by solving the finite element method, the average temperature T' of the heat-loaded surface is extracted. hot Calculate the temperature difference ΔT'=T' between the two surfaces. hot T' ref Subsequently, based on the one-dimensional Fourier heat conduction law, the overall equivalent thermal conductivity in this direction is calculated.

[0027]

[0028] Where H' is the characteristic dimension of the model in the computational direction; repeat the above process to calculate and obtain the equivalent thermal conductivity of the medium element in the X and Y directions. and Finally, a complete set of orthogonal anisotropic global equivalent thermophysical property parameters was obtained. , , .

[0029] The macroscopic structural model in step 7 is geometrically established based on the dimensions of the target structure. In terms of material definition, the target structure along the thickness direction, i.e., the Z-direction, is considered to be composed of J groups of periodically stacked ply units. Each ply unit consists of a material layer inheriting the properties of intermediate units and a virtual interface layer. The thermal properties of the material layers are directly assigned to the orthogonal anisotropic overall equivalent thermal properties output in step 6. , , The virtual interface layer attributes are still based on the equivalent thermal conductivity of the virtual interface layer in step 4. , , Configure the settings. Step 7, parametric damage simulation, uses a programmed script to simulate various typical damages, including interlayer debonding, matrix cracking, gap filler detachment, and surface ablation, on the macroscopic model. For simulating interlayer debonding damage, the model is used to locate the set of elements representing the virtual interface layer and switch their material properties from intact parameters to the thermal properties of air. For simulating cracking and gap filler detachment damage, the model is used to locate the set of elements corresponding to the damaged region and switch their material properties to the thermal properties of air. For simulating surface ablation damage, the model is used to modify the material properties of the elements in the corresponding damaged region.

[0030] The unit material properties modified in step 7 include equivalent thermal conductivity, specific heat capacity, and surface emissivity.

[0031] The parameterized damage simulation method in step 7 simulates damage by changing material properties without altering the model's geometry and mesh. The size, location, depth, and type of damage can all be flexibly defined and controlled through input parameters, thereby enabling the efficient generation of simulation models that simulate various types of damage.

[0032] Step 8, transient heat conduction simulation, simulates the unsteady heat exchange process experienced by the structure under the action of thermal load in a real external environment. The simulation sets the initial temperature field and time-varying boundary conditions according to the target working conditions. The boundary conditions include the convective heat transfer coefficient of the model surface, the ambient temperature, and the emissivity. The boundary conditions are set in segments to simulate the combined heat dissipation effect of convective heat transfer and thermal radiation of different intensities. The inner surface of the model is set according to the internal environment.

[0033] The conversion process in step 9 is as follows: First, based on the surface temperature field, material emissivity, and spectral response of the target camera, an ideal thermal radiation intensity image sequence is calculated according to Planck's blackbody radiation law; then, the key physical effects of a real infrared sensor are simulated sequentially for the ideal thermal radiation intensity image sequence: the point spread function and spatial resolution degradation of the optical system are simulated by Gaussian filtering; temporal random noise and spatial fixed pattern noise are added to simulate thermal noise and pixel response non-uniformity; finally, the continuous radiation intensity values ​​are quantized into digital grayscale values ​​with finite bit depth and sampled to the nominal pixel resolution of the camera, thereby outputting a high-fidelity simulated infrared image sequence.

[0034] The simulated infrared image sequence in step 9 retains the core thermal features caused by various types of damage while possessing the noise, blur, and non-uniform texture common to real infrared images, thus providing a high-quality image data foundation for the subsequent generation of standardized training samples.

[0035] The post-processing in step 10 involves automatically associating each simulated infrared image with and generating preset damage truth information to form a one-to-one corresponding sample pair. The damage truth information includes damage category labels and damage spatial characterization information. The damage spatial characterization information precisely defines the location, geometric shape, and size of the damage in the image.

[0036] Step 10, integration, refers to storing and managing all samples according to a preset, structured data organization format to generate a standardized damage infrared feature sample library. This standardized damage infrared feature sample library serves as the overall output of this method and is used to train and validate the intelligent identification model for surface structure damage.

[0037] The beneficial effects of this invention are that it provides a method for simulating and generating infrared features of aircraft surface structure damage for field detection, which combines high physical fidelity, high computational efficiency, and fully automated annotation capabilities. First, through the design of a "virtual interface layer" and "secondary homogenization," a significant improvement in the efficiency of full-scale transient thermal simulation is achieved while maintaining physical accuracy, making batch simulation of numerous damage conditions possible. Second, by simulating the sensor physical effects of an infrared camera with high fidelity, the simulated images are highly consistent with real field data in terms of noise, texture, and other features, effectively reducing the difference between the simulation and reality domains. Finally, by directly linking the simulation process with preset damage parameters, automatic, accurate, and zero-error generation of truth information is achieved, fundamentally solving the problems of high cost and poor consistency of manual annotation. This invention provides an efficient and reliable data generation method for training high-precision, highly robust intelligent field detection models, and has clear engineering practical value. Attached Figure Description

[0038] Figure 1 This is a flowchart of the simulation and sample generation method described in this invention.

[0039] Figure 2 This is a schematic diagram of step 2 of the present invention.

[0040] Figure 3 This is a schematic diagram of step 5 of the present invention.

[0041] Figure 4 This is a schematic diagram of step 7 of the present invention.

[0042] In the figure: 100, statistical representative volumetric unit model of needle-punched reinforced fiber web layer; 101, resin matrix; 102, Z-direction quartz needle-punched fiber; 103, quartz chopped fiber; 110, statistical representative volumetric unit model of needle-punched reinforced fiber fabric layer; 111, plain weave quartz fiber yarn structure; 200, medium unit model; 201, fiber web layer; 202, virtual interface layer; 203, fiber fabric layer; 204, sewing thread; 300, macroscopic finite element model; 301, material layer; 302, preset damage. Detailed Implementation

[0043] The present invention will be further described below with reference to the accompanying drawings: like Figure 1As shown, a method for simulating infrared features of surface structural damage on aircraft and generating samples is proposed. First, by fusing intrinsic material properties and manufacturing process parameters, a multi-scale physical simulation model from mesoscopic to macroscopic is constructed, and the concept of a virtual interface layer 202 is introduced to achieve efficient parameterized simulation of damage. Second, while ensuring the authenticity of the physical mechanism, a secondary homogenization strategy is used to significantly reduce computational complexity, thereby enabling transient heat conduction simulation of full-size structures and obtaining the real temperature field response of damage under external cooling conditions. Then, through high-fidelity infrared sensor imaging effect simulation, the temperature field is converted into a simulated infrared image that is highly consistent with the real acquired data in terms of noise, texture, and degradation characteristics. Finally, based on preset damage parameters, damage ground truth values ​​containing category labels and spatial representation information are automatically generated for the simulated infrared image, forming sample pairs, and a standardized damage infrared feature sample library is constructed in a structured manner.

[0044] Specifically Step 1: Obtaining multi-scale physical simulation parameters: Based on the process documents and material performance test results, obtain a set of multi-scale physical simulation parameters for the target thermal protection structure, providing a physical input benchmark for subsequent simulation modeling of heat conduction and damage from the microscopic to the macroscopic level; The parameter set includes basic material parameters and key process parameters; The basic material parameters include the thermal conductivity, specific heat capacity, and density of the resin matrix 101 and quartz fiber, as well as the overall surface emissivity of the composite material in the infrared band, which are used to define the material physical properties of models at all scales.

[0045] The key process parameters specifically include fiber reinforcement geometric parameters and Z-axis reinforcement process parameters: the fiber reinforcement geometric parameters cover the areal density, single-layer thickness, weaving structure, warp and weft yarn density, yarn cross-sectional shape and size, and fiber volume fraction of the fiber fabric layer 203, as well as the areal density, single-layer thickness, chopped fiber length range, and fiber volume fraction of the fiber web layer 201; the Z-axis reinforcement process parameters cover the needle punching density and needle punching depth of the needle punching process, and the sewing thread 204 material, diameter, pattern, and spacing of the sewing process; the key process parameters are used to accurately construct the geometric features of subsequent mesoscopic and mesoscopic scale models, wherein the needle punching density and needle punching depth directly determine the number and spatial distribution of needled fibers in the mesoscopic representative volume unit, and the sewing pattern and sewing spacing directly determine the three-dimensional geometric path and spatial density of the sewing thread 204 in the mesoscopic composite unit.

[0046] Step 2, as follows Figure 2 As shown, the construction of the meso-scale statistical representative volume unit is as follows: at the meso-scale, statistical representative volume unit models 110 for the needle-punched reinforced fiber fabric layer 203 and 100 for the needle-punched reinforced fiber web layer 201 are constructed respectively to calculate the equivalent thermophysical parameters of the two. The statistical representative volume unit model 110 of the needle-punched reinforced fiber fabric layer 203 consists of a resin matrix 101, a periodic plain-weave quartz fiber yarn structure 111 embedded in the resin matrix 101, and Z-direction quartz needle-punched fibers 102 randomly penetrating between the resin matrix 101 and the yarn. The statistical representative volume unit model of the needle-punched reinforced mesh consists of a resin matrix 101, quartz short-cut fibers 103 distributed in the resin matrix 101 at random positions and with random orientations, and Z-oriented quartz needle-punched fibers 102 that are also randomly distributed throughout.

[0047] The statistical representative volume element model uses tetrahedral elements for spatial discretization, and the material properties of each component are defined according to the basic material parameters of the parameter set obtained in step 1.

[0048] The thickness direction of the needle-punched reinforced fiber cloth layer 203, which represents the volumetric unit model 110, and the needle-punched reinforced fiber web layer 201, which represents the volumetric unit model 100, i.e., the Z-direction dimension, are determined according to the single-layer thickness parameters of the fiber cloth layer 203 and the fiber web layer 201 defined in step 1, so as to characterize a complete material single layer.

[0049] The dimension Lc of the needle-punched reinforcing fiber fabric layer 203, which statistically represents the volumetric unit model 110 in the in-plane (XY) direction, must simultaneously satisfy the requirements of periodic characterization of the weaving structure and statistical representativeness of the needle-punched fibers. Its calculation formula is as follows:

[0050] Wherein, P is the basic periodic dimension of the braided structure, in millimeters; k is the periodic quantity coefficient, a positive integer not less than 1, and its typical value range is 2 to 4 to ensure the stability of the equivalent properties; ρ is the needle density in the key process parameters obtained in step 1, in the number of needles per square millimeter; N is the number of needled fibers that need to be included in the statistical representative volume unit model 110 of the needled reinforcing fiber fabric layer 203, and the value range of N is 3 to 10 fibers to ensure statistical representativeness.

[0051] The dimension Lb of the statistical representative volume element model 100 in the in-plane direction of the needle-punched reinforced fiber web layer 201 is calculated based on the statistical representativeness requirements of the needle-punched fibers, and its calculation formula is as follows:

[0052] Wherein, ρ is the needle density in the key process parameters obtained in step 1, in units of needle count / square millimeter, and N' is the number of needled fibers that need to be included in the statistical representative volume unit model 100 of the needled reinforced fiber web 201. To ensure statistical representativeness, the value of N' ranges from 3 to 10 fibers.

[0053] The needle-punched reinforced fiber web layer 201 is represented by a statistical representative volume unit model 100. The short-cut fibers inside are realized using an equivalent modeling method: the small-diameter, randomly arranged quartz fiber filaments in the actual process are equivalent to a set of cylindrical rods with a uniform diameter of d millimeters and random distribution of length and spatial orientation under the statistical scale of this statistical representative volume unit model. This is strictly consistent with the fiber volume fraction defined in step 1. d is usually taken as 0.05-0.2 to solve the problem that it is difficult to directly and accurately model due to its complex real shape.

[0054] Step 3, Calculation of equivalent thermal properties: The equivalent thermal conductivity of the two statistical representative volume element models 203 and 201 constructed in Step 2 in the three principal directions of the material is obtained by finite element simulation calculation. The finite element simulation calculation is to perform independent steady-state thermal analysis in the three orthogonal directions of X, Y and Z for each statistical representative volume element model. The steady-state thermal analysis is performed sequentially for each specific direction. First, the equivalent thermal conductivity in the model thickness direction, defined as the Z-direction, is calculated. Then, a constant heat flux boundary condition with a constant heat flux density q and a constant temperature T are applied to two opposing surfaces characterizing this direction. ref Temperature boundary conditions, T ref Typically, 0℃ is used; the other four sides of the model are set as adiabatic boundary conditions, or periodic boundary conditions are applied to simulate the infinite periodic expansion of the material in the plane. After obtaining the steady-state temperature field, the average temperature T of the surface to which the heat flux boundary is applied is extracted. hot Calculate the average temperature difference ΔT = T between the two surfaces. hot T ref According to the one-dimensional Fourier law of heat conduction, this statistic represents the equivalent thermal conductivity k of the volume element model in the current direction, i.e., the Z-direction. z It can be calculated using the following formula:

[0055] Where H is the characteristic dimension (i.e., thickness) of the model in the calculation direction; repeat the above process to calculate and obtain the equivalent thermal conductivity k in the X and Y directions of the statistical representative volume element model. x With k y Finally, the equivalent thermal conductivity in the three principal directions of the needle-punched reinforced fiber fabric layer 203, representing the volume element model 110, is output. , , And the equivalent thermal conductivity of the three principal material directions of the statistical representative volume element model 100 of the needle-punched reinforced fiber web layer 201. , , The parameters constitute the key material properties for homogenizing microscopic heterostructures, providing input for subsequent heat transfer simulations at the mesoscopic and macroscopic scales.

[0056] Step 4, Virtual Interface Layer 202 Attribute Definition: Based on the mesoscopic equivalent thermophysical property parameters obtained in Step 3, define the virtual interface layer 202 for efficient simulation of interlayer damage, providing a basis for parameterized damage simulation in subsequent mesoscopic and macroscopic models; The virtual interface layer 202 is a conceptual thin layer, geometrically positioned between the needle-punched reinforcing fiber cloth layer 203 and the needle-punched reinforcing fiber mesh layer 201, used to characterize the resin-rich interlaminar transition region in the actual composite material. Regarding material properties, when simulating a perfectly intact structure, the equivalent thermal conductivity of the virtual interface layer 202 in the three principal material directions (X, Y, Z) is obtained by taking the arithmetic mean of the equivalent thermophysical properties of its adjacent fiber cloth layer 203 and fiber mesh layer 201 in that direction. The specific calculation formula is as follows:

[0057]

[0058]

[0059] in, , , These are the equivalent thermal conductivity coefficients of the virtual interface layer 202 in the three main material directions of X, Y, and Z, respectively.

[0060] When simulating interlayer debonding damage in a specific region, the thermal properties of the virtual interface layer 202 at the corresponding location can be switched to air properties, enabling physically realistic damage characterization without reconstructing the geometric model and computational mesh. This design effectively improves the modeling efficiency of multi-condition damage simulation.

[0061] Step 5, as follows Figure 3 As shown, the construction of the mesoscale intermediate unit and the explicit modeling of the sewing thread 204 are as follows: Based on the mesoscale equivalent thermal property parameters obtained in step 3 and the virtual interface layer 202 defined in step 4, a mesoscale intermediate unit model 200 is constructed. By integrating the mesoscale equivalent properties and sewing process features, a geometric and physical foundation is laid for the subsequent overall macroscopic homogenization calculation. The structure of the intermediate unit model 200 is formed by periodically stacking multiple basic ply units along the thickness direction, i.e., the Z direction. Each basic ply unit strictly follows a fixed sequence of "fiber cloth layer 203 - virtual interface layer 202 - fiber web layer 201 - virtual interface layer 202". The material thermophysical parameters of the fiber cloth layer 203 and the fiber web layer 201 are respectively assigned to the equivalent thermal conductivity of the needle-punched reinforced fiber cloth layer 203 obtained in step 3. , , Equivalent thermal conductivity of needle-punched reinforced fiber mesh ply 201 , , The material properties of the virtual interface layer 202 are based on those calculated in step 4. , , To assign.

[0062] The dimensions of the medium unit model 200 are determined according to the principle of representativeness: its in-plane (XY) dimensions must be greater than the sewing spacing defined in step 1 to ensure that the periodicity of the sewing pattern can be fully represented; its thickness (Z direction) dimension is composed of M of the above basic ply units stacked together to reasonably represent the thickness direction characteristics of the overall ply structure, while avoiding excessive complexity of the model. Generally, M can be 4-8.

[0063] In the medium unit model 200, the sewing thread 204 is explicitly geometrically modeled based on the sewing pattern and sewing spacing in the key process parameters obtained in step 1. The sewing thread 204 is established as a cylinder that runs through the entire thickness of the combined unit. Its spatial position and path are determined by the sewing pattern and sewing spacing, and its material properties are set according to the quartz fiber properties defined in step 1.

[0064] The intermediate unit model 200 integrates all the complex features inherited from the mesoscale, such as material anisotropy, interlayer interface properties, and Z-axis sewing reinforcement effect. Through a subsequent steady-state analysis, the originally complex multilayer heterogeneous structure can be homogenized into a single orthogonal anisotropic material unit. This key model dimensionality reduction and equivalence eliminates the need to model and calculate the underlying complex structure during subsequent full-scale transient analysis at the macroscale, thereby improving computational efficiency by orders of magnitude while ensuring physical accuracy. The underlying complex structure refers to the layer-by-layer geometry of the fiber cloth layer 203, the virtual interface layer 202, the fiber web layer 201, and the geometry of the sewing thread 204.

[0065] Step 6: Calculation of overall equivalent thermal properties of mesoscopic intermediate units: Perform steady-state thermal analysis on the mesoscopic intermediate units constructed in Step 5, calculate their overall equivalent thermal conductivity in the three principal material directions, and use this to homogenize the complex units into a set of orthogonal anisotropic macroscopic material parameters, and provide input for subsequent macroscopic simulation; realize the mesoscopic secondary homogenization after the mesoscopic homogenization.

[0066] The steady-state thermal analysis is performed independently along the three principal material directions (X, Y, Z) of the medium-sized element model 200. First, the overall equivalent thermal conductivity in the Z direction is calculated: a constant heat flux density q' boundary condition and a constant reference temperature T' are applied to the two opposing surfaces of the model in the Z direction (thickness direction). ref Boundary conditions, T' ref Typically, 0℃ is used; the other four sides are set as adiabatic boundary conditions; after obtaining the steady-state temperature field by solving the finite element method, the average temperature T' of the heat-loaded surface is extracted. hot Calculate the temperature difference ΔT'=T' between the two surfaces. hot T' ref Subsequently, based on the one-dimensional Fourier heat conduction law, the overall equivalent thermal conductivity in this direction is calculated.

[0067]

[0068] Where H' is the characteristic dimension of the model in the computational direction; repeat the above process to calculate and obtain the equivalent thermal conductivity of the medium element in the X and Y directions. and Finally, a complete set of orthogonal anisotropic global equivalent thermophysical property parameters was obtained. , , .

[0069] Step 7, as follows Figure 4 As shown, the macroscopic structural model construction and damage parameterization simulation are as follows: First, based on the overall equivalent thermophysical parameters obtained in step 6, a full-size macroscopic finite element model 300 of the target structure is established; then, by parametrically modifying the material properties, various types of preset damage 302 are efficiently simulated, providing a computational basis for subsequent transient thermal analysis.

[0070] The macroscopic structural model is geometrically established based on the dimensions of the target structure. In terms of material definition, the target structure along the thickness direction, i.e., the Z-direction, is considered to be composed of J groups of periodically stacked ply units. Each ply unit consists of a material layer 301 inheriting the properties of intermediate units and a virtual interface layer 202. The thermal properties of the material layer 301 are directly assigned to the orthogonal anisotropic overall equivalent thermal properties output in step 6. , , The properties of the virtual interface layer 202 are still based on the equivalent thermal conductivity of the virtual interface layer 202 in step 4. , , The model is configured in this way; the macroscopic model inherits the equivalent physical properties of the material without explicitly reconstructing the complex geometry of the underlying fibers, lay-ups and sewing thread 204, thereby significantly reducing the number of elements and computational complexity of the finite element model, creating conditions for subsequent rapid transient analysis.

[0071] The parametric damage simulation, through a programmed script, simulates various typical damages on the macroscopic model, including interlayer debonding, matrix cracking, gap filler detachment, and surface ablation. For the simulation of interlayer debonding damage, the element set representing the virtual interface layer 202 is located in the model, and its material properties are switched from intact state parameters to air thermal properties; this operation does not require changes to the model's geometry and mesh. For the simulation of cracking and gap filler detachment damage, the element set corresponding to the damaged region is located in the model, and its material properties are switched to air thermal properties. For the simulation of surface ablation damage, the element material properties of the corresponding damaged region are modified to simulate the degradation of material properties.

[0072] The modified unit material properties include equivalent thermal conductivity, specific heat capacity, and surface emissivity.

[0073] The parametric damage simulation method simulates damage by changing material properties without altering the model's geometry and mesh. The size, location, depth, and type of damage can all be flexibly defined and controlled through input parameters, thereby efficiently generating simulation models that simulate various types of damage.

[0074] Step 8, Transient heat conduction simulation and temperature field extraction: Based on the macroscopic structural model constructed in Step 7, a transient heat conduction simulation of the real external environment thermal load is performed to obtain the response of the structural surface under dynamic thermal environment and extract its time-series temperature field dataset for subsequent infrared image generation. The transient heat conduction simulation simulates the unsteady heat exchange process experienced by the structure under the action of thermal load in a real external environment. The simulation sets the initial temperature field and time-varying boundary conditions according to the target working conditions. The boundary conditions include the convective heat transfer coefficient of the model surface, the ambient temperature, and the emissivity. The boundary conditions are set in segments to simulate the combined heat dissipation effect of convective heat transfer and thermal radiation of different intensities. The inner surface of the model is set according to the internal environment.

[0075] Step 9, Infrared Image Generation and Sensor Effect Simulation: The time-series surface temperature field dataset extracted in Step 8 is converted into a high-fidelity simulated infrared image sequence. By simulating the imaging physical process of a real infrared camera, the simulated data is made consistent with the real field acquisition data in terms of features. The conversion process begins by calculating an ideal thermal radiation intensity image sequence based on the surface temperature field, material emissivity, and the spectral response of the target camera, according to Planck's blackbody radiation law. Then, the key physical effects of a real infrared sensor are simulated sequentially on this ideal thermal radiation intensity image sequence: Gaussian filtering is used to simulate the point spread function and spatial resolution degradation of the optical system; temporal random noise and spatial fixed pattern noise are added to simulate thermal noise and pixel response non-uniformity; finally, the continuous radiation intensity values ​​are quantized into digital grayscale values ​​with a finite bit depth and sampled to the camera's nominal pixel resolution, thereby outputting a high-fidelity simulated infrared image sequence.

[0076] The simulated infrared image sequence retains the core thermal features caused by various types of damage, while possessing the noise, blur, and non-uniform texture common to real infrared images, thus providing a high-quality image data foundation for the subsequent generation of standardized training samples.

[0077] Step 10: Generation of Standardized Damage Infrared Feature Sample Library: First, define a set of damage conditions containing different damage types, locations, and size parameters; based on the full-size macroscopic finite element model 300 of the target structure constructed in Step 7, according to the set of damage conditions, iteratively call the parametric damage simulation function in Step 7. In each iteration, based on a specific set of damage condition parameters, modify the material properties of the corresponding region of the model through a programmed script to efficiently generate a macroscopic finite element model 300 with a specific damage configuration; then, batch execute the transient heat conduction simulation described in Step 8 on the generated set of macroscopic finite element models 300 to obtain the time-series temperature field dataset corresponding to each model; then, batch execute the infrared image generation and sensor effect simulation process described in Step 9 on the time-series temperature field dataset to uniformly convert it into a simulated infrared image sequence; finally, perform automated post-processing and integration on the simulated infrared image sequence to construct a standardized damage infrared feature sample library that can be directly used for machine learning model training.

[0078] The post-processing involves automatically associating each simulated infrared image with and generating its preset damage truth information in step 7, forming a one-to-one corresponding sample pair; the damage truth information includes damage category labels and damage spatial characterization information.

[0079] The damage spatial characterization information should precisely define the location, geometry, and size of the damage in the image.

[0080] The integration refers to storing and managing all samples according to a preset, structured data organization format to generate a standardized damage infrared feature sample library; the standardized damage infrared feature sample library serves as the overall output of this method and is used to train and validate the intelligent recognition model for surface structure damage.

Claims

1. A method for simulating infrared features of surface structural damage on aircraft and generating samples, characterized in that, First, by integrating intrinsic material properties and manufacturing process parameters, a multi-scale physical simulation model is constructed, ranging from mesoscopic to macroscopic levels. A virtual interface layer concept is introduced to achieve efficient parametric simulation of damage. Second, while ensuring the authenticity of the physical mechanism, a secondary homogenization strategy significantly reduces computational complexity, enabling transient heat conduction simulation of full-size structures and obtaining the true temperature field response of damage under external cooling conditions. Then, through high-fidelity infrared sensor imaging effect simulation, the temperature field is converted into a simulated infrared image highly consistent with the actual acquired data in terms of noise, texture, and degradation characteristics. Finally, based on preset damage parameters, damage ground truth values ​​containing category labels and spatial representation information are automatically generated from the simulated infrared images, forming sample pairs, and a standardized damage infrared feature sample library is constructed in a structured manner.

2. The method for simulating infrared features of surface structural damage on aircraft and generating samples according to claim 1, characterized in that, The specific generation method is as follows: Step 1: Obtaining Multi-Scale Physical Simulation Parameters: Based on the process documents and material performance test results, obtain the set of multi-scale physical simulation parameters for the target thermal protection structure; Step 2: Construction of mesoscopic statistical representative volumetric unit: At the mesoscopic scale, statistical representative volumetric unit models of needle-punched reinforced fiber fabric layer and needle-punched reinforced fiber web layer are constructed respectively to calculate the equivalent thermophysical parameters of the two. Step 3, Calculation of equivalent thermal properties: The equivalent thermal conductivity of the two statistical representative volume element models, needle-punched reinforced fiber cloth and needle-punched reinforced fiber web, constructed in Step 2, is obtained in the three principal directions of the material through finite element simulation. The finite element simulation is to perform independent steady-state thermal analysis in the three orthogonal directions of X, Y, and Z for each statistical representative volume element model. Step 4, Virtual Interface Layer Attribute Definition: Based on the mesoscopic equivalent thermophysical parameters obtained in Step 3, define a virtual interface layer for efficient simulation of interlayer damage; Step 5, Mesoscale intermediate unit construction and explicit modeling of sewing thread: Based on the mesoscale equivalent thermal property parameters obtained in Step 3 and the virtual interface layer defined in Step 4, a mesoscale intermediate unit model is constructed. By integrating mesoscale equivalent properties and sewing process features, a geometric and physical foundation is laid for subsequent overall macroscopic homogenization calculations. Step 6: Calculation of overall equivalent thermal properties of mesoscopic intermediate units: Perform steady-state thermal analysis on the mesoscopic intermediate units constructed in Step 5, and calculate their overall equivalent thermal conductivity in the three main material directions. This is used to homogenize the complex units into a set of orthogonal anisotropic macroscopic material parameters and to provide input for subsequent macroscopic simulation. Achieving secondary homogenization at the mesoscale after achieving homogenization at the mesoscale. Step 7, Macroscopic Structural Model Construction and Damage Parametric Simulation: First, based on the overall equivalent thermophysical parameters obtained in Step 6, a full-size macroscopic finite element model of the target structure is established; then, by parametrically modifying material properties, various types of preset damage are efficiently simulated, providing a computational basis for subsequent transient thermal analysis. Step 8, Transient heat conduction simulation and temperature field extraction: Based on the macroscopic structural model constructed in Step 7, a transient heat conduction simulation of the real external environment thermal load is performed to obtain the response of the structural surface under dynamic thermal environment and extract its time-series temperature field dataset for subsequent infrared image generation. Step 9, Infrared Image Generation and Sensor Effect Simulation: The time-series surface temperature field dataset extracted in Step 8 is converted into a high-fidelity simulated infrared image sequence. By simulating the imaging physical process of a real infrared camera, the simulated data is made consistent with the real field acquisition data in terms of features. Step 10: Generation of Standardized Damage Infrared Feature Sample Library: First, define a set of damage conditions containing different damage types, locations, and size parameters; based on the full-size macroscopic finite element model of the target structure constructed in Step 7, according to the set of damage conditions, iteratively call the parametric damage simulation function in Step 7. In each iteration, based on a specific set of damage condition parameters, modify the material properties of the corresponding region of the model through a programmed script to efficiently generate a macroscopic finite element model with a specific damage configuration; then, batch execute the transient heat conduction simulation described in Step 8 on the generated set of macroscopic finite element models to obtain the time-series temperature field dataset corresponding to each model; then, batch execute the infrared image generation and sensor effect simulation process described in Step 9 on the time-series temperature field dataset to uniformly convert it into a simulated infrared image sequence; finally, perform automated post-processing and integration on the simulated infrared image sequence to construct a standardized damage infrared feature sample library that can be directly used for machine learning model training.

3. The method for simulating infrared features of surface structural damage on aircraft and generating samples according to claim 2, characterized in that, The parameter set in step 1 includes basic material parameters and key process parameters. The basic material parameters include the thermal conductivity, specific heat capacity, and density of the resin matrix and quartz fiber, as well as the overall surface emissivity of the composite material in the infrared band, used to define the material physical properties of models at all scales. The key process parameters specifically include fiber reinforcement geometric parameters and Z-axis reinforcement process parameters. The fiber reinforcement geometric parameters cover the areal density, single-layer thickness, weaving structure, warp and weft yarn density, yarn cross-sectional shape and size, and fiber volume fraction of the fiber fabric layer, as well as the areal density, single-layer thickness, chopped fiber length range, and fiber volume fraction of the fiber web layer. The Z-axis reinforcement process parameters cover the needle punching density and needle punching depth of the needle punching process, as well as the sewing thread material, diameter, pattern, and spacing of the sewing process.

4. The method for simulating infrared features of surface structural damage on aircraft and generating samples according to claim 2, characterized in that, The statistical representative volumetric unit model of the needle-punched reinforced fiber fabric layer in step 2 consists of a resin matrix, a periodic plain-weave quartz fiber yarn structure embedded in the resin matrix, and Z-direction quartz needle-punched fibers randomly penetrating between the resin matrix and the yarn.

5. The method for simulating infrared features of surface structural damage on aircraft and generating samples according to claim 2, characterized in that, The statistical representative volumetric unit model of the needle-punched reinforced mesh in step 2 consists of a resin matrix, quartz short-cut fibers distributed in the resin matrix at random positions and orientations, and Z-direction quartz needle-punched fibers that are also randomly distributed throughout.

6. The method for simulating infrared features of surface structural damage on aircraft and generating samples according to claim 2, characterized in that, The micro-statistical representative volume element model in step 2 uses tetrahedral elements for spatial discretization, and the material properties of each component are defined according to the basic material parameters of the parameter set obtained in step 1.

7. The method for simulating infrared features of surface structural damage on aircraft and generating samples according to claim 2, characterized in that, The thickness direction, i.e. the Z-axis dimension, of the statistical representative volume unit model of the needle-punched reinforced fiber cloth layer and the statistical representative volume unit model of the needle-punched reinforced fiber web layer in step 2 is determined according to the single-layer thickness parameters of the fiber cloth layer and the fiber web layer defined in step 1, so as to characterize a complete single layer of material.

8. The method for simulating infrared features of surface structural damage on aircraft and generating samples according to claim 2, characterized in that, In step 2, the representative volume element model of the needle-punched reinforced fiber fabric layer has an in-plane dimension L in the XY direction. c It needs to simultaneously meet the requirements of periodic characterization of the braided structure and statistical representativeness of the needle-punched fibers. The calculation formula is as follows: Wherein, P is the basic periodic dimension of the braided structure, in millimeters; k is the periodic quantity coefficient, a positive integer not less than 1, and its typical value range is 2 to 4 to ensure the stability of the equivalent properties; ρ is the needle density in the key process parameters obtained in step 1, in the number of needles per square millimeter; N is the number of needled fibers that need to be included in the statistical representative volume unit model of the needled reinforced fiber fabric layer, and the value range of N is 3 to 10 fibers to ensure statistical representativeness.

9. The method for simulating infrared features of surface structural damage on aircraft and generating samples according to claim 2, characterized in that, The dimension L of the in-plane direction of the representative volume element model of the needle-punched reinforced fiber mesh layer in step 2 is described in step 2. b The main consideration is to meet the statistical representativeness requirements of needle-punched fibers, and the calculation formula is as follows: Wherein, ρ is the needle density in the key process parameters obtained in step 1, in units of needles per square millimeter, and N' is the number of needled fibers that need to be included in the statistical representative volume unit model of the needled reinforced fiber web. To ensure statistical representativeness, the value of N' ranges from 3 to 10 fibers.

10. The method for simulating infrared features of surface structural damage on an aircraft and generating samples according to claim 2, characterized in that, The statistical representative volume unit model of the needle-punched reinforced fiber web in step 2 is implemented using an equivalent modeling method for the chopped fibers inside. In the actual process, the small-diameter, randomly arranged quartz fiber filaments are equivalent to a set of cylindrical rods with a uniform diameter of d millimeters and random distribution of length and spatial orientation under the statistical scale of this statistical representative volume unit model. This model strictly maintains consistency with the fiber volume fraction defined in step 1. d is usually taken as 0.05-0.2 to solve the problem that it is difficult to directly and accurately model due to its complex real shape.

11. The method for simulating infrared features of surface structural damage on aircraft and generating samples according to claim 2, characterized in that, Step 3, steady-state thermal analysis, is performed sequentially for each specific direction. First, the equivalent thermal conductivity in the model thickness direction, defined as the Z-direction, is calculated. Then, a constant heat flux density q and a constant temperature T are applied to two opposing surfaces characterizing this direction. ref Temperature boundary conditions, T ref Typically, 0℃ is used; the other four sides of the model are set as adiabatic boundary conditions, or periodic boundary conditions are applied to simulate the infinite periodic expansion of the material in the plane. After obtaining the steady-state temperature field, the average temperature T of the surface to which the heat flux boundary is applied is extracted. hot Calculate the average temperature difference ΔT = T between the two surfaces. hot T ref According to the one-dimensional Fourier law of heat conduction, this statistic represents the equivalent thermal conductivity k of the volume element model in the current direction, i.e., the Z-direction. z It can be calculated using the following formula: Where H is the characteristic dimension of the model in the calculation direction; repeat the above process to calculate and obtain the equivalent thermal conductivity k in the X and Y directions of the statistical representative volume element model. x With k y Finally, the equivalent thermal conductivity in the three principal directions of the statistical representative volume element model of the needle-punched reinforced fiber fabric is output. , , The equivalent thermal conductivity in the three principal directions of the statistical representative volume element model of the needle-punched reinforced fiber web layer. , , The parameters constitute the key material properties for homogenizing microscopic heterostructures, providing input for subsequent heat transfer simulations at the mesoscopic and macroscopic scales.

12. The method for simulating infrared features of surface structural damage on an aircraft and generating samples according to claim 11, characterized in that, The virtual interface layer in step 4 is a conceptual thin layer, and its geometric position is defined between the needle-punched reinforced fiber cloth layer and the needle-punched reinforced fiber mesh layer. Regarding material properties, when simulating a structurally intact state, the equivalent thermal conductivity of the virtual interface layer in the three principal material directions (X, Y, and Z) is obtained by taking the arithmetic mean of the equivalent thermophysical properties of its adjacent fiber cloth layer and fiber mesh layer in that direction. The specific calculation formula is as follows: in, , , These are the equivalent thermal conductivity coefficients of the virtual interface layer in the three main material directions of X, Y, and Z, respectively.

13. The method for simulating infrared features of surface structural damage on an aircraft and generating samples according to claim 12, characterized in that, The structure of the unit model in step 5 is formed by periodically stacking multiple basic ply units along the thickness direction, i.e., the Z-direction. Each basic ply unit strictly follows a fixed sequence of "fiber cloth layer - virtual interface layer - fiber web layer - virtual interface layer". The material thermophysical parameters of the fiber cloth layer and the fiber web layer respectively impart the equivalent thermal conductivity of the needle-punched reinforced fiber cloth layer. , , Equivalent thermal conductivity of needle-punched reinforced fiber mesh ply , , The material properties of the virtual interface layer are based on... , , To assign.

14. The method for simulating infrared features of surface structural damage on an aircraft and generating samples according to claim 3, characterized in that, In step 5, the size of the unit model is determined according to the principle of representativeness: its in-plane size, i.e., the XY direction, needs to be greater than the stitching spacing to ensure that the periodicity of the stitching pattern can be fully represented; its thickness, i.e., the Z direction dimension, is composed of M of the above basic ply units stacked together to reasonably represent the thickness direction characteristics of the overall ply structure, while avoiding excessive complexity of the model. Generally, M can be 4-8.

15. The method for simulating infrared features of surface structural damage on an aircraft and generating samples according to claim 3, characterized in that, In step 5, the unit model is used to perform explicit three-dimensional geometric modeling of the sewing thread based on the sewing pattern and sewing spacing in the key process parameters obtained. The sewing thread is established as a cylinder that runs through the thickness of the entire combined unit, and its spatial position and path are determined by the sewing pattern and sewing spacing.

16. The method for simulating infrared features of surface structural damage on an aircraft and generating samples according to claim 2, characterized in that, Step 6, steady-state thermal analysis, involves performing independent simulations sequentially along the three principal material directions (X, Y, Z) of the medium-sized element model. First, the overall equivalent thermal conductivity in the Z direction is calculated: a constant heat flux density q' boundary condition and a constant reference temperature T' are applied to the two opposing surfaces of the model in the Z direction (thickness direction). ref Boundary conditions, T' ref Typically, 0℃ is used; the other four sides are set as adiabatic boundary conditions; after obtaining the steady-state temperature field by solving the finite element method, the average temperature T' of the heat-loaded surface is extracted. hot Calculate the temperature difference ΔT'=T' between the two surfaces. hot T' ref Subsequently, based on the one-dimensional Fourier heat conduction law, the overall equivalent thermal conductivity in this direction is calculated. Where H' is the characteristic dimension of the model in the computational direction; repeat the above process to calculate and obtain the equivalent thermal conductivity of the medium element in the X and Y directions. and Finally, a complete set of orthogonal anisotropic global equivalent thermophysical property parameters was obtained. , , .

17. The method for simulating infrared features of surface structural damage on an aircraft and generating samples according to claim 12, characterized in that, The macroscopic structural model in step 7 is geometrically established based on the dimensions of the target structure. In terms of material definition, the target structure along the thickness direction, i.e., the Z-direction, is considered to be composed of J groups of periodically stacked ply units. Each ply unit consists of a material layer inheriting the properties of intermediate units and a virtual interface layer. The thermal properties of the material layers are directly assigned to the orthogonal anisotropic overall equivalent thermal properties output in step 6. , , The virtual interface layer attributes are still based on the equivalent thermal conductivity of the virtual interface layer in step 4. , , Configure the settings.

18. The method for simulating infrared features of surface structural damage on an aircraft and generating samples according to claim 2, characterized in that, Step 7, damage parameterization simulation, is performed using a programmed script to simulate various typical damage types on the macroscopic model, including interlayer debonding, matrix cracking, gap filling detachment, and surface ablation. For the simulation of interlayer debonding damage, the model is used to locate the set of elements representing the virtual interface layer and switch their material properties from intact parameters to the thermal properties of air. For the simulation of cracking and gap filling detachment damage, the model is used to locate the set of elements corresponding to the damaged region and switch their material properties to the thermal properties of air. For the simulation of surface ablation damage, the model is used to modify the material properties of the elements in the corresponding damaged region.

19. The method for simulating infrared features of surface structural damage on an aircraft and generating samples according to claim 18, characterized in that, The modified unit material properties include equivalent thermal conductivity, specific heat capacity, and surface emissivity.

20. The method for simulating infrared features of surface structural damage on an aircraft and generating samples according to claim 18, characterized in that, The damage parameterization simulation method in step 7 simulates damage by changing material properties without altering the model's geometry and mesh. The size, location, depth, and type of damage can all be flexibly defined and controlled through input parameters, thereby efficiently generating simulation models that simulate various types of damage.

21. The method for simulating infrared features of surface structural damage on an aircraft and generating samples according to claim 2, characterized in that, Step 8, transient heat conduction simulation, simulates the unsteady heat exchange process experienced by the structure under the action of thermal load in a real external environment. The simulation sets the initial temperature field and time-varying boundary conditions according to the target working conditions. The boundary conditions include the convective heat transfer coefficient of the model surface, the ambient temperature, and the emissivity. The boundary conditions are set in segments to simulate the combined heat dissipation effect of convective heat transfer and thermal radiation of different intensities. The inner surface of the model is set according to the internal environment.

22. The method for simulating infrared features of surface structural damage on an aircraft and generating samples according to claim 2, characterized in that, The conversion process in step 9 is as follows: First, based on the surface temperature field, material emissivity, and spectral response of the target camera, an ideal thermal radiation intensity image sequence is calculated according to Planck's blackbody radiation law; then, the key physical effects of a real infrared sensor are simulated sequentially for the ideal thermal radiation intensity image sequence: the point spread function and spatial resolution degradation of the optical system are simulated by Gaussian filtering; temporal random noise and spatial fixed pattern noise are added to simulate thermal noise and pixel response non-uniformity; finally, the continuous radiation intensity values ​​are quantized into digital grayscale values ​​with finite bit depth and sampled to the nominal pixel resolution of the camera, thereby outputting a high-fidelity simulated infrared image sequence.

23. The method for simulating infrared features of surface structural damage on an aircraft and generating samples according to claim 2, characterized in that, The simulated infrared image sequence in step 9 retains the core thermal features caused by various types of damage while possessing the noise, blur, and non-uniform texture common to real infrared images, thus providing a high-quality image data foundation for the subsequent generation of standardized training samples.

24. The method for simulating infrared features of surface structural damage on an aircraft and generating samples according to claim 2, characterized in that, The post-processing in step 10 involves automatically associating and generating damage truth information for each simulated infrared image, forming a one-to-one corresponding sample pair. The damage truth information includes damage category labels and damage spatial characterization information. The damage spatial characterization information precisely defines the location, geometric shape, and size of the damage in the image.

25. The method for simulating infrared features of surface structural damage on an aircraft and generating samples according to claim 2, characterized in that, Step 10, integration, refers to storing and managing all samples according to a preset, structured data organization format to generate a standardized damage infrared feature sample library. This standardized damage infrared feature sample library serves as the overall output of this method and is used to train and validate the intelligent identification model for surface structure damage.

Images