Method and system for detecting internal defects of composite parts

Abstract

The application provides a composite material part internal defect detection method and a detection system. The composite material part internal defect detection method comprises the following steps: obtaining test data of a simulation part, and constructing a model training data set. The model training data set is input into an initial machine learning model for training to obtain a defect prediction model. A heat source is applied to a composite material part to be detected, and a thermal imaging image of a shooting surface of the composite material part to be detected is obtained. The thermal imaging image is segmented into multiple image regions. Pixel point coordinates, pixel point temperature time sequence information of each image region, and geometric feature information of the shooting surface of the composite material part to be detected are obtained. The pixel point coordinates, the pixel point temperature time sequence information of each image region, and the geometric feature information of the composite material part to be detected are input into the defect prediction model to obtain defect information of each image region. The detection method can improve the detection accuracy and efficiency.

Description

Technical Field

[0001] This invention relates to the detection of defects in composite materials, and more particularly to a method and system for detecting internal defects in composite material parts. Background Technology

[0002] Composite materials are frequently used in modern aero-engines, such as in fan blades, high-pressure turbine outer rings, and guide vanes. Composite materials are characterized by low density and sufficiently high strength. They are typically categorized into resin-based composites and ceramic-based composites. However, the manufacturing process of composite materials introduces numerous macroscopic and microscopic defects. For example, resin-based materials may exhibit internal defects such as bubbles, pores, delamination, and fiber breakage. Ceramic-based composites primarily suffer from internal defects including cracks, delamination, inclusions, porosity, and residual silicon. These defects are difficult to completely avoid, are dispersed, have diverse causes, complex mechanisms, and significantly different impacts on component strength, making accurate quantification challenging. Surface defects are relatively easy to detect using methods such as microscopic visual inspection and fluorescence penetrant testing. However, internal defects are difficult to detect without damaging the material. Traditional non-destructive testing methods for internal defects primarily utilize industrial computed tomography (CT) scanning technology. CT scanning, often referred to as industrial CT, is a non-destructive testing technique used for non-destructive testing (NDT) and non-destructive evaluation (NDE). It uses X-rays to pass through the object being tested and generate detailed images of its internal structure, thereby detecting information such as material defects, component integrity, internal structural features, and geometric dimensions. However, industrial computed tomography (CT) scans require high-energy X-rays, which are typically expensive and have long testing cycles.

[0003] Under the current level of composite material industry research and development, the yield rate of composite material blades for aero-engines is not high, and there are many defects, which may affect the safety of aero-engines. Non-destructive testing of the prepared materials is usually required, that is, defect detection without damaging the parts. Composite material parts, such as blades, are often complex thin-walled structures. Conventional thermal imaging inspection technology involves heating one side of the thin-walled structure of the composite material part using thermal radiation or heat conduction, while simultaneously using an infrared camera on the front to record the temperature in real time.

[0004] However, due to the presence of defects such as pores, cracks, and inclusions within thin plates, the thermal conductivity varies, resulting in different spatiotemporal patterns in thermal imaging. Conventional thermal imaging techniques rely on manual observation of thermal images for judgment, making it difficult to discern the number, size, and location of defects that may exist within the thin plate, and are prone to missed or false detections. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for detecting internal defects in composite material parts, which can improve detection accuracy and efficiency.

[0006] One aspect of the present invention provides a method for detecting internal defects in composite material parts, comprising: acquiring test data of a simulated part and constructing a model training dataset; inputting the model training dataset into an initial machine learning model for training to obtain a defect prediction model; applying a heat source to the composite material part to be tested and acquiring a thermal imaging image of the photographed surface of the composite material part to be tested; segmenting the thermal imaging image into multiple image regions; acquiring pixel coordinates, pixel temperature time-series information, and geometric feature information of the photographed surface of the composite material part to be tested for each image region; inputting the pixel coordinates, pixel temperature time-series information, and geometric feature information of the composite material part to be tested into the defect prediction model to obtain defect information for each image region.

[0007] In one embodiment, acquiring test data of the simulated component and constructing a model training dataset includes: selecting parameter features of the simulated component, establishing a simulated component model, performing finite element heat transfer analysis on the simulated component model, and constructing a model training dataset based on the analysis results of the finite element heat transfer analysis.

[0008] In one embodiment, the step of selecting the parameter characteristics of the simulated part, establishing a simulated part model, and performing finite element heat transfer analysis on the simulated part model includes: selecting a simulated part with the same material characteristics as the composite material part to be tested; performing parametric modeling on simulated parts with different parameter characteristics to obtain an initial model of the simulated part; and adding defects with different parameter characteristics inside the initial model of the simulated part to obtain a simulated part model.

[0009] In one embodiment, the step of establishing a simulation model and performing finite element heat transfer analysis on the simulation model further includes: dividing the simulation model into finite element meshes and changing the application position of the heat source to carry out finite element heat transfer analysis.

[0010] In one embodiment, the step of constructing a model training dataset based on the analysis results of finite element heat transfer analysis includes: extracting the analysis results of finite element heat transfer analysis, selecting the node coordinates and node temperature time series information of each node in the analysis results as original features, and constructing an initial dataset; performing Pearson correlation analysis on the initial dataset to reduce the number of original features in the initial dataset, transforming the features of the initial dataset through principal component analysis to obtain derived features; and adding the derived features to the initial dataset as a model training dataset.

[0011] In one embodiment, applying a heat source to the composite material part to be tested and acquiring a thermal imaging image of the composite material part to be tested includes: applying a heat source to one side of the composite material part to be tested; using the side opposite the heat source application side of the composite material part to be tested as the imaging side, and acquiring a thermal imaging image of the imaging side using an infrared camera.

[0012] In one embodiment, the thermal imaging image is segmented into multiple image regions, wherein the size of the image region is greater than three times the size of the internal defect.

[0013] In one embodiment, acquiring the pixel coordinates, pixel temperature time-series information, and geometric feature information of the imaging surface of the composite material part to be tested in each of the image regions includes: acquiring the pixel coordinates and pixel temperature time-series information of each of the image regions; acquiring the spatial curvature and corresponding normal thickness of each node of the imaging surface of the digital model of the composite material part to be tested based on the digital model of the part; obtaining the pixel coordinates of the imaging surface of the composite material part to be tested by affine transformation of the three-dimensional coordinates of the node set of the imaging surface of the digital model of the part, thereby obtaining the spatial curvature and corresponding normal thickness of the imaging surface of the composite material part to be tested.

[0014] In one embodiment, the step of inputting the pixel coordinates, pixel temperature time-series information, and geometric feature information of the composite material to be tested into the defect prediction model to obtain defect information of each image region includes: inputting the pixel coordinates, pixel temperature time-series information, spatial curvature, and corresponding normal thickness of each image region of the imaging surface of the composite material part to be tested into the defect prediction model to obtain the defect size and location of each image region.

[0015] Another aspect of the present invention provides a detection system for internal defects in composite material parts, comprising: a heat source; an infrared camera for capturing thermal images; a worktable for placing the composite material part to be detected and the infrared camera; and a processing device; wherein the processing device includes a memory and a processor, the processor being connected to the memory and configured to implement the detection method for internal defects in composite material parts as described in any of the above embodiments.

[0016] The present invention provides a method for detecting internal defects in composite material parts. Based on the time-series temperature information of the parts' pixels, the method uses a machine learning model to predict defects in thermal imaging images. This method overcomes the shortcomings of traditional thermal imaging technology, which is difficult to detect or produces ambiguous results. It can fully utilize the information of camera pixels to capture minute temperature differences, avoiding the uncertainty that may be caused by manual judgment in traditional thermal imaging. This method can detect potential defects more conveniently and efficiently, with more accurate defect detection results and lower costs compared to manual observation. It can achieve low-cost, high-efficiency non-destructive testing of defects. Attached Figure Description

[0017] The above and other features, properties and advantages of the present invention will become more apparent from the following description taken in conjunction with the accompanying drawings and embodiments, wherein:

[0018] Figure 1 It is a detection system for internal defects in composite material parts before improvement;

[0019] Figure 2 This is a schematic diagram of an embodiment of a detection system for internal defects in composite material parts according to the present invention;

[0020] Figure 3 This is a schematic flowchart of an embodiment of a method for detecting internal defects in composite material parts according to the present invention;

[0021] Figure 4 This is a schematic flowchart of another embodiment of the method for detecting internal defects in composite material parts according to the present invention;

[0022] Figure 5 yes Figure 3 A schematic diagram of automated modeling of the simulated part in the detection method shown;

[0023] Figure 6 yes Figure 3 A schematic diagram of step S500 in the detection method shown. Detailed Implementation

[0024] Figure 1 The original system for detecting internal defects in composite material parts is shown. Figure 1 As shown, conventional thermal imaging detection technology involves heating one side of the composite material part 11 with a heat source 1 (thermal radiation or thermal conduction), while an infrared camera 2 records the temperature in real time on the front side (i.e. the side opposite to the surface to which the heat source is applied) to form an image 12.

[0025] However, due to the presence of defects such as pores, cracks, and inclusions inside the composite material part 11 (e.g., thin plate), the thermal conductivity varies, resulting in different spatiotemporal variation patterns in thermal imaging. Conventional thermal imaging technology relies on manual observation of thermal images for judgment, making it difficult to distinguish the number, size, and location of defects that may exist inside the thin plate, and it is prone to missed or incorrect detections.

[0026] Reference will now be made in detail to embodiments of the invention, one or more examples of which are illustrated in the accompanying drawings. Each example is provided to explain the invention and not to limit it. In fact, it will be apparent to those skilled in the art that various modifications and variations may be made to the invention without departing from the scope or spirit thereof. For example, a feature shown or described as part of one embodiment may be used with another embodiment to produce yet another embodiment. Therefore, the invention is intended to cover these modifications and variations that fall within the scope of the appended claims and their equivalents.

[0027] The term "infrared thermography" is a non-destructive detection technique that generates thermal images, temperature distribution maps, and temperature rise profiles by detecting the amount of infrared radiation emitted by an object. This technology can be used to monitor and diagnose potential problems in many different fields, including building energy efficiency, electrical fault detection, manufacturing process monitoring, and medical applications.

[0028] Figure 2 An embodiment of the detection system for internal defects in composite material parts according to the present invention is shown. The detection system for internal defects in composite material parts according to the present invention includes a worktable 55, an infrared camera 2, a heat source 1, and a processing device 3. The worktable 55 is used to place the composite material part 33 to be inspected and the infrared camera 2; the infrared camera 2 is fixed to the worktable 55 by a bracket 54. The composite material part 33 to be inspected is a composite material part 33 that is prone to defects during the manufacturing process, and may include flat plates, geometric specimens with complex features, such as most thin-walled aero-engine blade parts 33 (e.g.,...). Figure 2 (The airfoil thin-walled structure shown).

[0029] Heat source 1 is used to heat the composite material part 33 to be tested. Heat source 1 is applied to one side of the composite material part 33 using heat radiation or heat conduction; this side serves as the application surface of heat source 1. The part should be kept as still as possible during the testing process. A quartz lamp, gas, or similar material can be used as heat source 1. The application point should be as close as possible to the surface of the composite material part 33. The power of heat source 1 needs to meet certain requirements, but it must not exceed the working temperature of the composite material to prevent damage.

[0030] While heat source 1 is applied, an infrared camera 2 is used on the opposite side of the surface where heat source 1 is applied (i.e., the imaging surface) to record the temperature field changes over time in real time, thus capturing thermal images. The infrared camera 2 records each frame of image information after the application of heat source 1 begins, obtaining video information about temperature changes. The thermal imaging video recorded by the infrared camera 2 must maintain a frame rate of at least 100Hz and a resolution of at least 1080P to ensure the accuracy of subsequent defect prediction.

[0031] The processing device 3 includes a memory and a processor, the processor being connected to the memory and configured to implement the method for detecting internal defects in composite material parts according to the present invention. The processing device 3 may be implemented using one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), processors, controllers, microcontrollers, microprocessors, other electronic devices for performing the above functions, or selected combinations of the above devices.

[0032] Figure 3 A schematic flowchart of an embodiment of the method for detecting internal defects in composite material parts according to the present invention is shown. Figure 3 As shown, the method for detecting internal defects in composite material parts according to the present invention includes steps S100 to S600:

[0033] In step S100, test data of the simulated parts are obtained to construct a model training dataset.

[0034] In step S200, the model training dataset is input into the initial machine learning model for training to obtain the defect prediction model.

[0035] In step S300, a heat source is applied to the composite material part to be inspected, and a thermal imaging image of the photographed surface of the composite material part to be inspected is acquired.

[0036] In step S400, the thermal imaging image is segmented into multiple image regions. (Reference) Figure 2 The thermal imaging image 34 is segmented to form multiple image regions 35.

[0037] In step S500, the pixel coordinates, pixel temperature time sequence information, and geometric feature information of the imaging surface of the composite material part to be detected are obtained for each image region.

[0038] In step S600, the pixel coordinates, pixel temperature time sequence information, and geometric feature information of the composite material to be detected in each image region are input into the defect prediction model to obtain the defect information of each image region.

[0039] Compared to traditional thermal imaging techniques that rely on visual inspection to identify defects, the detection method of this invention can identify defects by using a defect prediction model based on the temporal changes in temperature reflected by internal defects in the part.

[0040] The present invention provides a method for detecting internal defects in composite material parts. Based on the time-series temperature information of the parts' pixels, the method uses a machine learning model to predict defects in thermal imaging images. This method overcomes the shortcomings of traditional thermal imaging technology, which is difficult to detect or produces ambiguous results. It can fully utilize the information of camera pixels to capture minute temperature differences, avoiding the uncertainty that may be caused by manual judgment in traditional thermal imaging. This method can detect potential defects more conveniently and efficiently, with more accurate defect detection results and lower costs compared to manual observation. It can achieve low-cost, high-efficiency non-destructive testing of defects.

[0041] The detection method of the present invention can be applied to composite material parts of aero-engines, and can also be applied to the detection of internal defects in parts made of other materials.

[0042] In one embodiment, step S100 further includes steps S110 to S120:

[0043] In step S110, the parameter characteristics of the simulated component are selected, a simulated component model is established, and a finite element heat transfer analysis is performed on the simulated component model.

[0044] In step S120, a model training dataset is constructed based on the analysis results of the finite element heat transfer analysis.

[0045] In this embodiment, steps S110 to S120 of the present invention use low-cost automatic parametric finite element calculation results instead of a large amount of test sample data, eliminating the need for complicated sample testing to collect data, greatly reducing the cost of acquiring datasets for AI model training, and also increasing efficiency.

[0046] Further, step S110 includes steps S111 to S114:

[0047] In step S111, a simulated part with material characteristics consistent with the composite material part to be tested is selected.

[0048] In step S112, parametric modeling is performed on the simulated parts with different parameter characteristics to obtain the initial model of the simulated parts. The simulated parts can be selected as flat plates, and their parameter characteristics are as follows: Figure 5 As shown, it includes length L, width W, and height H. The "simulation parts with different parameter characteristics" in step S112 refer to flat simulation parts with different lengths, widths, and heights. These parameter characteristics can be modeled using CAD software such as UG.

[0049] In step S113, defects with different parameter features are added inside the initial model of the simulated part to obtain the simulated part model. Step S113 means that defects of different sizes are automatically created and added at different locations inside the initial model of the simulated part. The defects can be ellipsoidal, and the parameter features of the embedded ellipsoidal defects include the minor axis a, major axis b, and the spatial coordinates of the ellipsoid's center (x0, y0, z0) and the Euler angles (α, β, γ) of the ellipsoid, such as... Figure 5 As shown.

[0050] In step S114, the simulated model is meshed using finite element methods, and the application position of the heat source is changed to perform finite element heat transfer analysis. In S114, finite element software is used to parameterize the heat source position, automatically mesh, and finally perform finite element heat transfer calculations.

[0051] Step S120 further includes steps S121 to S123:

[0052] In step S121, the analysis results of the finite element heat transfer analysis are extracted, and the node coordinates and node temperature time series information of each node in the analysis results are selected as original features to construct an initial dataset. Representative information is selected from the analysis results of the finite element heat transfer analysis and the time series output from the historical detection data set to select input data for the machine learning model.

[0053] The temporal information extracted from the finite element calculation results includes the node coordinates Ni(x, y, z) (i is the index of all nodes, x, y, z are the image pixel coordinates), and the temporal information of temperature for each node Ni(t, Temp) (i is the index of all nodes, t is the time, and Temp is the temperature value of the corresponding coordinate) as input data for the machine learning model.

[0054] The initial dataset can also be built based on a large amount of actual inspection historical data (including information such as whether the part defects are qualified or unqualified).

[0055] In step S122, Pearson correlation analysis is performed on the initial dataset to reduce the number of original features. Principal component analysis (PCA) is then used to transform the features of the initial dataset to obtain derived features. Pearson correlation analysis can be performed to reduce the number of features, and principal component analysis (PCA) can be applied to transform the features (e.g., spatial perspective transformation, affine transformation, first derivative of time, second derivative of time, square root, power operation, exponential form, trigonometric transformation) to derive new features from the original features, thereby enhancing the modeling capability of the machine learning algorithm.

[0056] The spatial curvature of each node is calculated. The plate thickness corresponding to the normal at each node, the length L, width W, and height H of the simulated plate are all input to the machine learning model. The defect minor axis a, major axis b, spatial coordinates (x0, y0, z0), and spatial Euler angles (α, β, γ) are used as outputs. Principal component analysis (PCA) is applied to transform the features, performing perspective and affine transformations on the spatial coordinates (x0, y0, z0), and performing first and second time derivative transformations on the node temperature time series information Ni(t, Temp).

[0057] In step S123, the derived features are added to the initial dataset as the model training dataset. The transformed and derived new features can also be added to the training to enhance the modeling ability of the machine learning algorithm.

[0058] Using low-cost, automated parametric finite element calculation results instead of a large amount of experimental sample data eliminates the need for cumbersome sample testing and data collection, greatly reducing the cost of acquiring datasets for AI model training and increasing efficiency.

[0059] In S200, the initial machine learning model can use machine learning models such as decision trees, SVM, random forests, and neural networks to train and test datasets in different feature spaces (original, reduced, transformed, and derived) to build a defect prediction model that can predict the depth, size, and location of defects based on the spatiotemporal changes in the temperature field of the structural surface.

[0060] The detection method of this invention uses an AI model for defect prediction, which can make full use of the camera pixel information, capture minute temperature differences, avoid the uncertainty that may be caused by manual judgment in traditional thermal imaging, and make the defect detection results more accurate.

[0061] The detection method of this invention targets the global location of the composite material part to be inspected, and combines it with a low-cost artificial intelligence method, resulting in a more powerful defect prediction capability. It can predict information such as the size and spatial location of cracks.

[0062] In one embodiment, step S300 further includes steps S310 to S320:

[0063] In step S310, a heat source is applied to one side of the composite material part to be tested.

[0064] In step S320, the opposite surface of the heat source application surface of the composite material part to be tested is used as the imaging surface, and a thermal imaging image of the imaging surface is acquired by an infrared camera.

[0065] In related technologies, the heat source heats the imaging surface, and the imaging records the cooling process of the sample surface after it has been heated by uniform spatial light pulses from a flash array. However, in the detection method of this invention, the surface heated by the heat source and the surface recorded by the thermal imaging are different relative surfaces, thus recording the sample's heating process and providing more stable temperature change information.

[0066] Considering the large variation in surface curvature of the entire composite material part under test, it does not perfectly match the simulated part. Therefore, it is conceivable to segment the surface via step S400. When the segmentation is sufficiently small, the result for each image region is comparable to the simulated part used in the training process described above.

[0067] In step S400, the image region cannot be divided too small, otherwise the effect of temperature changes caused by defects cannot be identified. Therefore, the size of each segmented image region is at least larger than the size of the internal defect to at least cover the internal defect. Preferably, the size of each segmented image region is greater than three times the size of the internal defect, where size represents area. The size of the internal defect can be given by a range based on common manufacturing processes. That is, different composite materials correspond to a general range of internal defect sizes, and the size of the image region can be determined based on this range of internal defect sizes.

[0068] In one embodiment, in step S500, the geometric feature information of the imaged surface of the composite material part to be inspected is acquired through a point cloud registration system. The point cloud registration system is a system used to align multiple point cloud datasets to a unified coordinate system.

[0069] refer to Figure 6 Based on the above embodiments, step S500 further includes steps S510 to S530:

[0070] In step S510, the pixel coordinates and pixel temperature time sequence information of each image region are obtained.

[0071] In step S520, based on the digital model of the composite material part to be tested, the spatial curvature and corresponding normal thickness of each node of the imaging surface of the digital model of the part are obtained.

[0072] In step S530, the three-dimensional coordinates of the node set of the imaging surface of the digital model of the part are obtained by affine transformation to obtain the pixel coordinates of the imaging surface of the composite material part to be inspected, thereby obtaining the spatial curvature and the corresponding normal thickness of the imaging surface of the composite material part to be inspected.

[0073] Furthermore, in S510, the set of pixels for acquiring the shooting surface information is N(x,y), and the recorded image information is represented as Zone1: pixel coordinates Nn(x,y), pixel temperature time-series information Nn(t,Temp), Zone2: pixel coordinates Nm(x,y), pixel temperature time-series information Nm(t,Temp), and so on. Here, Zone is the zone number, N is the number of pixels within each zone, t is the time, x and y are the image pixel coordinates, and Temp is the temperature value at the corresponding coordinates.

[0074] By introducing the temporal variation of temperature at the same pixel (i.e., the same spatial location), the traditional thermal infrared imaging system overcomes the challenges of reduced detection speed due to excessively high resolution and missed or false detections due to excessively low resolution. By incorporating time-varying features into existing spatial thermal imaging, more efficient and accurate detection can be achieved at lower resolutions.

[0075] In S520, the digital model of the composite material part to be inspected can be obtained from the finite element model before its processing, or by scanning the point cloud with a 3D scanner, remodeling, and then obtaining the digital model of the part. The spatial curvature at each node of the image surface of the digital model of the part, as well as the corresponding normal thickness at each node, are obtained through CAD or CAE. This node set is a three-dimensional coordinate system N'(x,y,z).

[0076] In S530, the set of shooting surface nodes N'(x,y,z) in the digital model of the part is transformed into the set of shooting surface information pixels N(x,y) through affine transformation. Therefore, the curvature information and thickness information can be obtained under the actual shooting surface information pixel set N(x,y).

[0077] In one embodiment, step S600 involves inputting the pixel coordinates, pixel temperature time-series information, spatial curvature of the image surface, and corresponding normal thickness of each image region of the composite material part to be inspected into the defect prediction model to obtain the defect size and location of each image region. Specifically, the pixel coordinates Nn(x, y), pixel temperature time-series information Nn(t, Temp), and geometric shape information of the actual part's image surface in each region are jointly input into the defect prediction model for defect prediction, and finally, the potential crack size and location of each region are output.

[0078] Because of defects such as pores and cracks inside parts, the thermal conductivity varies. By training a defect prediction model at low cost and analyzing the changes in thermal imaging in real time, information such as the number, size, and location of possible defects inside the part can be derived.

[0079] In conclusion, Figure 4Another embodiment of the detection method of the present invention is shown. The detection method of the present invention uses a program to automatically read the results of thermal imaging pixels and combines them with an AI model to predict defects. It can predict the depth, size and location of defects based on the changes in the surface temperature field of the part. It solves the problems of how to achieve rapid and low-cost detection of internal defects in composite material parts with heterogeneous and complex structures and how to achieve high-efficiency and accurate detection of internal defects in composite material parts. It avoids the drawbacks of traditional thermal imaging technology, such as difficulty in detection or fuzzy results, and can detect possible defects more conveniently and efficiently.

[0080] The detection method of this invention can detect defects at greater depths than traditional thermal imaging, making it suitable for parts with complex geometries. Traditional thermal imaging technology, on the other hand, requires defects to be close to the imaging surface or pass through an interface layer, and is often used for flat specimens; otherwise, the thermal imaging image cannot distinguish the effect of the defect on temperature conduction.

[0081] The detection method of this invention is universal and can be extended to other imaging analysis techniques based on other principles, such as ultrasound imaging, which can identify based on the feedback echo information.

[0082] While the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the invention. Any variations and modifications can be made by those skilled in the art without departing from the spirit and scope of the invention. Therefore, any modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention, without departing from the scope of the invention, fall within the protection scope defined by the claims of the present invention.

Claims

1. A method for detecting internal defects in composite material parts, characterized in that, include: Obtain test data of the simulated parts and construct a model training dataset; The training dataset of the model is input into the initial machine learning model for training to obtain the defect prediction model; A heat source is applied to the composite material part to be inspected, and a thermal imaging image of the photographed surface of the composite material part to be inspected is acquired; The thermal imaging image is divided into multiple image regions; Obtain the pixel coordinates, pixel temperature time sequence information, and geometric feature information of the image surface of the composite material part to be detected in each image region; The pixel coordinates, pixel temperature time-series information, and geometric feature information of the composite material to be detected in each image region are input into the defect prediction model to obtain the defect information of each image region.

2. The detection method as described in claim 1, characterized in that, The process of acquiring experimental data from the simulated components and constructing a model training dataset includes: Select the parameter characteristics of the simulated component, establish a model of the simulated component, and perform finite element heat transfer analysis on the simulated component model; Based on the analysis results of the finite element heat transfer analysis, a model training dataset was constructed.

3. The detection method as described in claim 2, characterized in that, The process of selecting parameter characteristics of the simulated component, establishing a simulated component model, and performing finite element heat transfer analysis on the simulated component model includes: Select a simulated part with material characteristics consistent with the composite material part to be tested; Parametric modeling is performed on simulated parts with different parameter characteristics to obtain the initial model of the simulated parts; By adding defects with different parameter features to the interior of the initial model of the simulation part, a simulation part model is obtained.

4. The detection method as described in claim 3, characterized in that, The process of establishing a simulation model and performing finite element heat transfer analysis on the simulation model further includes: The simulated model was meshed using finite element methods, and the location of the heat source was changed to perform finite element heat transfer analysis.

5. The detection method as described in claim 2, characterized in that, The model training dataset is constructed based on the analysis results of the finite element heat transfer analysis, including: Extract the analysis results of the finite element heat transfer analysis, select the node coordinates and node temperature time series information of each node in the analysis results as the original features, and construct the initial dataset; Pearson correlation analysis is performed on the initial dataset to reduce the number of original features in the initial dataset, and principal component analysis is used to transform the features of the initial dataset to obtain derived features; The derived features are added to the initial dataset as the model training dataset.

6. The detection method according to any one of claims 1-5, characterized in that, The process of applying a heat source to the composite material part to be inspected and acquiring a thermal imaging image of the composite material part to be inspected includes: Apply a heat source to one side of the composite material part to be inspected; The opposite surface of the heat source application surface of the composite material part to be tested is used as the imaging surface, and a thermal imaging image of the imaging surface is acquired by an infrared camera.

7. The detection method according to any one of claims 1-5, characterized in that, The thermal imaging image is divided into multiple image regions, wherein the size of each image region is greater than three times the size of the internal defect.

8. The detection method according to any one of claims 1-5, characterized in that, The acquisition of pixel coordinates, pixel temperature time-series information, and geometric feature information of the image surface of the composite material part to be detected for each image region includes: Obtain the pixel coordinates and pixel temperature time-series information for each of the image regions; Based on the digital model of the composite material part to be tested, the spatial curvature and corresponding normal thickness of each node of the imaging surface of the digital model are obtained. The three-dimensional coordinates of the node set of the photographed surface of the digital model of the part are obtained by affine transformation to obtain the pixel coordinates of the photographed surface of the composite material part to be tested, thereby obtaining the spatial curvature and the corresponding normal thickness of the photographed surface of the composite material part to be tested.

9. The detection method as described in claim 8, characterized in that, The step of inputting the pixel coordinates, pixel temperature time-series information, and geometric feature information of the composite material to be detected into the defect prediction model to obtain defect information for each image region includes: The pixel coordinates, pixel temperature time-series information, spatial curvature, and corresponding normal thickness of each image region of the image surface of the composite material part to be inspected are input into the defect prediction model to obtain the defect size and location of each image region.

10. A system for detecting internal defects in composite material parts, characterized in that, include: Heat source; Infrared cameras are used to capture thermal images; A workbench is used to place the composite material parts to be inspected and the infrared camera. as well as Processing device; wherein, The processing device includes a memory and a processor connected to the memory and configured to implement a method for detecting internal defects in composite material parts as described in any one of claims 1-9.

Images