A non-destructive testing method and system for a ceramic matrix composite flame tube

Abstract

The present application relates to the technical field of nondestructive testing, and more particularly to a nondestructive testing method and system for a ceramic matrix composite flame tube, the method comprising: constructing a three-dimensional geometric model of the flame tube and a waveguide propagation model of the ceramic matrix composite; selecting an appropriate wave mode, dynamically adjusting an excitation frequency and an incident angle to obtain an ultrasonic signal of the flame tube; obtaining a two-dimensional ultrasonic image set; using a semantic segmentation tool to perform pixel labeling to generate a corresponding segmentation mask image; obtaining samples according to the segmentation mask image; constructing a defect generation model based on a convolutional neural network, inputting the samples into the defect generation model for training and reconstruction to obtain defect features; generating a three-dimensional imaging data image to obtain a defect detection result; and the system comprising a processor, a phased array probe, a mechanical arm, and a human-computer interaction display screen. The present application can improve defect recognition accuracy and detection efficiency, and meet the quality control requirements of high-performance components of an aero-engine.

Description

Technical Field

[0001] This invention relates to the technical field of nondestructive testing, and more specifically, to a nondestructive testing method and system for ceramic matrix composite flame tubes. Background Technology

[0002] Ceramic matrix composites (CMCs) are widely used in high-temperature components such as the flame tubes of aero engines due to their excellent high-temperature stability, low density, and corrosion resistance. However, the complex fiber structure and geometry of CMC flame tubes make them prone to micropores, cracks, and delamination defects. These defects can lead to structural failure under high-temperature and high-load operation, threatening the safety of the engine.

[0003] Existing non-destructive testing methods include X-ray computed tomography (CT), which, while enabling three-dimensional imaging, suffers from low resolution for low-density defects and is time-consuming. Infrared thermal imaging, limited by its depth, can only identify surface and near-surface defects, failing to cover the full thickness of the flame tube wall and thus making it difficult to detect internal defects. Traditional ultrasonic testing, a common method for CMC component inspection, is limited by the high acoustic attenuation characteristics of CMC materials and scattering caused by fiber structures, making it difficult to identify minute defects. Furthermore, it has poor adaptability to complex geometries such as the annular surface and cooling holes of the flame tube, resulting in low coverage. Consequently, the waveform and excitation parameters of traditional ultrasonic probes are difficult to dynamically adapt to material properties, making it difficult to meet the high-precision and high-efficiency quality control requirements of aero-engines. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of existing non-destructive testing technologies in meeting the testing requirements of ceramic matrix composite flame tubes, and to provide a non-destructive testing method and system for ceramic matrix composite flame tubes, which can improve defect identification accuracy, improve testing efficiency, and meet the quality control requirements of high-performance components of aero-engines.

[0005] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: A non-destructive testing method for ceramic matrix composite flame tubes is provided, comprising the following steps: S1. Construct a three-dimensional geometric model of the flame tube and a waveguide propagation model of ceramic matrix composite material; S2. Based on the three-dimensional geometric model and waveguide propagation model, select suitable wave patterns for different regions of the flame tube geometry, and dynamically adjust the excitation frequency and incident angle according to the regional characteristics corresponding to the regional geometry to obtain the ultrasonic signal of the flame tube. S3. The acquired ultrasonic signal from the flame tube is amplified, filtered, and then converted into a two-dimensional ultrasonic image set; S4. Use a semantic segmentation tool to annotate the pixels of the two-dimensional ultrasound image set to generate a corresponding segmentation mask image; decompose the segmentation mask image into multiple image blocks and classify the image blocks to obtain normal samples and abnormal samples; S5. Construct a defect generation model based on a convolutional neural network, and input the normal samples and abnormal samples into the defect generation model for training and reconstruction to obtain defect features; S6. Stack the two-dimensional ultrasound image set that has undergone the above processing to generate a three-dimensional imaging data image, and obtain the defect detection result based on the defect features therein.

[0006] This invention discloses a non-destructive testing method for a ceramic matrix composite flame tube. By selecting appropriate waveforms for different regions of the flame tube's geometry and dynamically adjusting the excitation frequency and incident angle based on the corresponding regional characteristics, the method ensures clear signal reception even in the high acoustic attenuation environment of the ceramic matrix composite material, thus improving the accuracy of the acquired ultrasonic signals from the flame tube. Furthermore, by using a semantic segmentation tool to generate corresponding segmentation mask images from a two-dimensional ultrasonic image set, and then decomposing the segmentation mask images into multiple image blocks, the method ensures that each image block can capture the smallest defect features of the flame tube, thereby improving the accuracy of defect detection. Finally, by inputting normal and abnormal samples into a defect generation model for training and reconstruction to obtain defect features, the method enables precise defect identification, improves defect recognition accuracy and detection efficiency, and meets the quality control requirements of high-performance components for aero-engines.

[0007] Further, step S1 includes the following steps: S11. Obtain the regional geometry and corresponding regional characteristics of the flame tube by laser scanning to construct the three-dimensional geometric model; wherein, the regional geometry of the flame tube includes a straight wall region, a curved transition region, and a cooling hole array region; the regional characteristics include the diameter and wall thickness of the straight wall region, the radius of curvature of the curved transition region, and the diameter and spacing of each cooling hole in the cooling hole array region; S12. Construct the waveguide propagation model based on the acoustic properties of the ceramic matrix composite material; wherein, the acoustic properties of the ceramic matrix composite material include sound velocity, attenuation coefficient, and anisotropy coefficient.

[0008] Further, step S2 includes: for the straight wall region: selecting a Lamb wave and setting a first excitation frequency; for the curved transition region: selecting a Lamb wave and setting a second excitation frequency; for the cooling hole array region: selecting a surface wave and setting a third excitation frequency; wherein the second excitation frequency is greater than the first excitation frequency, and the first excitation frequency is greater than the third excitation frequency; and the incident angle is dynamically adjusted according to the curvature of the region geometry and the sound velocity ratio of the ceramic matrix composite material.

[0009] Further, step S3 includes: using a signal amplifier to increase the gain of the acquired ultrasonic signal from the flame tube to a preset multiple, then performing filtering processing through a wavelet transform algorithm, and then converting it into a two-dimensional ultrasonic image set; wherein, the filtering processing includes: filtering out low-frequency interference through a wavelet transform algorithm and extracting the effective frequency band signal.

[0010] Further, step S4 includes the following steps: S41. The two-dimensional ultrasound image set is pixel-annotated using a semantic segmentation tool. The pixel annotation includes background pixels, defect-free foreground pixels, and defective pixels. Then, a corresponding segmentation mask image is generated. S42. The segmentation mask image is decomposed into multiple image blocks according to a certain pixel size, wherein each image block may contain any one or more pixel labels; S43. Remove some image blocks that contain only background pixels, so that the remaining image blocks have a corresponding proportion of background-defect-free foreground pixels; S44. Classify the remaining image blocks: Image blocks containing defect-free foreground pixels but not defective pixels are classified as normal samples; image blocks containing defective pixels are classified as abnormal samples.

[0011] Further, in step S44, for an image block containing defective pixels: The image blocks belonging to the straight wall area and curved transition area of ​​the flame tube are decomposed into multiple image blocks according to a certain pixel size, with the geometric center of the defect pixel as the reference. The image block belonging to the cooling hole array area of ​​the flame tube is expanded to the corresponding hole wall part with the geometric center of the defect pixel as the reference, and then decomposed into multiple image blocks according to a certain pixel size. Then the abnormal samples are obtained.

[0012] Further, step S5 includes the following steps: S51. Construct a defect generation model based on convolutional neural networks; S52. Input the normal samples into the defect generation model, train it with the goal of minimizing the reconstruction error between the normal samples and the model-generated samples, and form generated samples; compare the normal samples with the generated samples and adjust the reconstruction accuracy of the normal flame tube structure; S53. Input both the normal samples and the abnormal samples into the defect generation model for reconstruction, and calculate the abnormal score using pixel-level mean square error; S54. Set an anomaly score threshold according to the defect type, and determine the corresponding defect type for samples that exceed the corresponding anomaly score threshold.

[0013] Further, in step S52, the key indicators of the normal sample and the generated sample are compared to ensure that the reconstruction accuracy of the normal flame tube structure is ≥95%; wherein, the key indicators include: the roundness deviation of the cooling hole array area of ​​the rocket tube is <5%, the gray standard deviation of the hole wall part of the cooling hole array area of ​​the rocket tube is <10, and the gray value deviation of the circumferential uniformity of the straight wall area of ​​the rocket tube is <10. In step S54, the defect types and their corresponding abnormality score thresholds include: porosity defects with a threshold of 0.06; crack defects with a threshold of 0.08; and defects around cooling holes with a threshold of 0.10.

[0014] Furthermore, the non-destructive testing method for the ceramic matrix composite flame tube further includes step S7: verifying and calibrating the defect generation model using a standard ceramic matrix composite specimen; wherein step S7 includes the following steps: S71. Take a standard ceramic matrix composite specimen, wherein the standard ceramic matrix composite specimen includes pore defects, crack defects, and defects around cooling holes; S72. Use the trained defect generation model to detect the standard ceramic matrix composite specimen, generate the resolution and positioning accuracy of the corresponding three-dimensional imaging data image, and obtain the first defect detection result; S73. If the resolution and positioning accuracy obtained in step S72 reach the set threshold, the verification is complete; S74. The defects of the standard ceramic matrix composite specimen are calibrated by X-ray to obtain a second defect detection result; the second defect detection result is compared with the first defect detection result obtained in step S72. If the defect identification rate is confirmed to be ≥97%, the calibration is completed.

[0015] The present invention also provides a non-destructive testing system applied to the above-mentioned non-destructive testing method. The non-destructive testing system includes a processor, and further includes a phased array probe, a robotic arm, and a human-machine interface display screen that are communicatively connected to the processor. The phased array probe is connected to the drive end of the robotic arm.

[0016] Compared with the prior art, the beneficial effects of the present invention are: This invention discloses a non-destructive testing method and system for ceramic matrix composite flame tubes. By selecting appropriate waveforms for different regions of the flame tube geometry and dynamically adjusting the excitation frequency and incident angle according to the regional characteristics corresponding to the regional geometry, it ensures clear signal reception under high acoustic attenuation conditions in ceramic matrix composites, improving the accuracy of the acquired ultrasonic signals from the flame tube. Furthermore, by using semantic segmentation tools to generate corresponding segmentation mask images from a two-dimensional ultrasonic image set, and then decomposing the segmentation mask images into multiple image blocks, the image blocks can capture the minimum defect features of the flame tube, improving the accuracy of defect detection. Finally, by inputting normal and abnormal samples into a defect generation model for training and reconstruction to obtain defect features, it enables precise defect determination, improves defect identification accuracy, enhances detection efficiency, and meets the quality control requirements of high-performance components for aero-engines. Attached Figure Description

[0017] Figure 1 This is a flowchart of a non-destructive testing method for a ceramic matrix composite flame tube according to the present invention; Figure 2 This is a schematic diagram of the structure of a non-destructive testing system according to the present invention.

[0018] In the attached diagram: 100, processor; 200, phased array probe; 300, robotic arm; 400, human-machine interface display screen; 500, flame tube. Detailed Implementation

[0019] The present invention will be further described below with reference to specific embodiments. The accompanying drawings are for illustrative purposes only, representing schematic diagrams rather than actual physical objects, and should not be construed as limiting the scope of this patent. To better illustrate the embodiments of the present invention, some components in the drawings may be omitted, enlarged, or reduced, and do not represent the actual dimensions of the product. It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.

[0020] In the accompanying drawings of the embodiments of the present invention, the same or similar reference numerals correspond to the same or similar components. In the description of the present invention, it should be understood that if terms such as "upper," "lower," "left," "right," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, they are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, the terms used to describe positional relationships in the drawings are only for illustrative purposes and should not be construed as limiting the present patent. For those skilled in the art, the specific meaning of the above terms can be understood according to the specific circumstances.

[0021] Example 1 like Figure 1The first embodiment of the non-destructive testing method for a ceramic matrix composite flame tube according to the present invention is shown, which includes the following steps: S1. Construct a three-dimensional geometric model of the flame tube 500, and construct a waveguide propagation model of ceramic matrix composite material; S2. Based on the three-dimensional geometric model and waveguide propagation model, select suitable wave patterns for different regions of the flame tube 500, and dynamically adjust the excitation frequency and incident angle according to the regional characteristics corresponding to the regional geometry to obtain the ultrasonic signal of the flame tube 500. S3. The acquired ultrasonic signal from the flame tube 500 is amplified, filtered, and then converted into a two-dimensional ultrasonic image set; S4. Use a semantic segmentation tool to annotate pixels in the two-dimensional ultrasound image set to generate corresponding segmentation mask images; decompose the segmentation mask images into multiple image blocks and classify the image blocks to obtain normal samples and abnormal samples; S5. Construct a defect generation model based on a convolutional neural network, input normal samples and abnormal samples into the defect generation model for training and reconstruction, and obtain defect features; S6. Stack the two-dimensional ultrasound image set that has undergone the above processing to generate a three-dimensional imaging data image, and obtain the defect detection result based on the defect features therein.

[0022] This invention discloses a non-destructive testing method for a ceramic matrix composite flame tube. By selecting appropriate waveforms for different regions of the flame tube's geometry and dynamically adjusting the excitation frequency and incident angle based on the corresponding regional characteristics, the method ensures clear signal reception even in the high acoustic attenuation environment of the ceramic matrix composite material, thus improving the accuracy of the acquired ultrasonic signals from the flame tube. Furthermore, by using a semantic segmentation tool to generate corresponding segmentation mask images from a two-dimensional ultrasonic image set, and then decomposing the segmentation mask images into multiple image blocks, the method ensures that each image block can capture the smallest defect features of the flame tube, thereby improving the accuracy of defect detection. Finally, by inputting normal and abnormal samples into a defect generation model for training and reconstruction to obtain defect features, the method enables precise defect identification, improves defect recognition accuracy and detection efficiency, and meets the quality control requirements of high-performance components for aero-engines.

[0023] Example 2 This embodiment is a second embodiment of a non-destructive testing method for a ceramic matrix composite flame tube. This embodiment is similar to the first embodiment, except that step S1 includes the following steps: S11. The regional geometry and corresponding regional characteristics of the flame tube 500 are obtained by laser scanning to construct a three-dimensional geometric model; wherein, the regional geometry of the flame tube 500 includes a straight wall region, a curved transition region, and a cooling hole array region; the regional characteristics include the diameter and wall thickness of the straight wall region, the radius of curvature of the curved transition region, and the diameter and spacing of each cooling hole in the cooling hole array region; Specifically, a laser scanner with a resolution of ≤0.03mm was selected to scan the flame tube 500; S12. Construct a waveguide propagation model based on the acoustic properties of ceramic matrix composites; wherein, the acoustic properties of ceramic matrix composites include sound velocity, attenuation coefficient, and anisotropy coefficient. Specifically, a waveguide propagation model is constructed based on a sound velocity of 5800 m / s, an attenuation coefficient of 1.8 dB / cm, and an anisotropy coefficient of ±15%.

[0024] In this embodiment, step S2 includes: for the straight wall region: selecting a Lamb wave and setting a first excitation frequency; for the curved transition region: selecting a Lamb wave and setting a second excitation frequency; for the cooling hole array region: selecting a surface wave and setting a third excitation frequency; wherein, the second excitation frequency is greater than the first excitation frequency, and the first excitation frequency is greater than the third excitation frequency; and the incident angle is dynamically adjusted according to the curvature of the region geometry and the sound velocity ratio of the ceramic matrix composite material. Specifically, the first excitation frequency, the second excitation frequency, and the third excitation frequency can be set to 4MHz, 6MHz, and 2MHz, respectively. A 4MHz Lamb wave is selected in the straight wall region to enhance the reflected signal from the pore defects by utilizing its standing wave characteristics. A 6MHz Lamb wave is selected in the curved transition region to adapt to the waveform changes caused by curvature. A 2MHz surface wave is selected in the cooling hole array region to avoid scattering interference from the hole walls and ensure clear signal under the high acoustic attenuation environment of CMC. The incident angle is maintained at 25-35°.

[0025] In this embodiment, step S3 includes: amplifying the acquired ultrasonic signal from the flame tube 500 to a preset multiple using a signal amplifier, then filtering it using a wavelet transform algorithm, and finally converting it into a two-dimensional ultrasonic image set; wherein, the filtering process includes: filtering out low-frequency interference (<1.8MHz) using a wavelet transform algorithm and extracting the effective frequency band signal of 1.8-8MHz; it should be noted that the effective frequency band signal refers to the concentrated frequency band of the reflected wave signals from typical defects such as pores and cracks in the flame tube 500.

[0026] In this embodiment, step S4 includes the following steps: S41. Use a semantic segmentation tool to perform pixel annotation on the two-dimensional ultrasound image set. The pixel annotation includes background pixels, defect-free foreground pixels, and defect pixels. Then, generate the corresponding segmentation mask image. Specifically, each two-dimensional ultrasound image in the set was segmented using a semantic segmentation tool. For defective areas, the threshold grayscale value was >120, while for defect-free areas, the threshold was between 80 and 120. Pixel annotation was also performed. It should be noted that background pixels refer to the ultrasound ineffective area, defect-free foreground pixels refer to the ultrasound effective area of ​​a normal flame tube structure, and defective pixels refer to the ultrasound effective area of ​​an abnormal flame tube structure. Furthermore, the segmentation masking process involved using Labelme annotation software to mark defective regions on each image in the two-dimensional ultrasound image set, providing a basis for subsequent image patch selection and sample classification. S42. Decompose the segmentation mask image into multiple image blocks according to a certain pixel size, wherein each image block may contain any one or more pixel labels; Specifically, the segmented mask image is decomposed into fixed-size image blocks of 128×128 pixels, and these pixel-sized image blocks are capable of carrying the features of the flame tube 500's minimum defect of 0.1mm. S43. Remove some image blocks that contain only background pixels, so that the remaining image blocks have an appropriate ratio of background to defect-free foreground pixels; Specifically, removing image blocks that contain only background pixels can be used to filter out some invalid data. In this embodiment, removing 70% or more of the image blocks that contain only background pixels, leaving image blocks with a 20% to 30% ratio of background-defect-free foreground pixels, can be used to assist the subsequent defect generation model in learning the feature distribution of background and defect-free foreground pixels. It should be noted that image blocks containing background-defect-free foreground pixels refer to pixel blocks that simultaneously contain background pixels and defect-free foreground pixels. S44. Classify the remaining image patches: Image patches containing defect-free foreground pixels but not defective pixels are classified as normal samples, which can be used to train the defect generation model to recognize the features of the defect-free part of the flame tube 500; image patches containing defective pixels are classified as abnormal samples, which can be used to train the defect generation model to recognize the features of the defective part of the flame tube 500 and the interference features of the cooling hole wall; among them, for image patches containing defective pixels: The image blocks belonging to the straight wall area and curved transition area of ​​the flame tube 500 are decomposed into multiple image blocks of 128×128 pixels, with the geometric center of the defect pixel as the reference. At the same time, the three-dimensional coordinate position information of each pixel block is obtained with the annular circumferential angle of the flame tube 500 as the X-axis and the axial depth as the Y-axis. The image block belonging to the cooling hole array area of ​​the flame tube 500 is taken as the reference and extended by 5mm to the corresponding hole wall part, and then decomposed into multiple image blocks of 128×128 pixel size. At the same time, the three-dimensional coordinate position information of each pixel block can also be obtained. It should be noted that extending 5mm to the corresponding hole wall part means extending 5mm along the extension direction of the cooling hole depth, that is, extending 5mm along the Y-axis direction. Then the abnormal samples are obtained.

[0027] In this embodiment, step S5 includes the following steps: S51. Construct a defect generation model based on convolutional neural networks; S52. Input normal samples into the defect generation model and train it with the goal of minimizing the reconstruction error between normal samples and model-generated samples. Learn the core features of normal flame tube structure and form generated samples. Then, compare normal samples with generated samples and adjust the reconstruction accuracy of normal flame tube structure. Specifically, the key indicators of normal samples and generated samples are compared to ensure that the reconstruction accuracy of normal flame tube structure is ≥95%. Among them, the key indicators include: the roundness deviation of the cooling hole array area of ​​the rocket tube is <5%, the gray standard deviation of the hole wall part of the cooling hole array area of ​​the rocket tube is <10, and the gray value deviation of the circumferential uniformity of the straight wall area of ​​the rocket tube is <10. S53. Input both normal and abnormal samples into the defect generation model for reconstruction, and calculate the abnormal score using pixel-level mean square error (MSE); It should be noted that, due to the significant difference in grayscale values ​​between the defect area and the normal area of ​​the ceramic matrix composite flame tube 500, specifically, the fluctuation range of the defect area is 3-5 times that of the normal area. Therefore, the reconstruction error of abnormal samples is significantly higher than that of normal samples, which can effectively distinguish between normal and abnormal samples. S54. Set an anomaly score threshold according to the defect type, and determine the corresponding defect type for samples that exceed the corresponding anomaly score threshold; Specifically, the defect types and their corresponding anomaly score thresholds include: pore defects located in the straight wall region with a threshold of 0.06; crack defects located in the curved transition region with a threshold of 0.08; and defects around the cooling holes located in the cooling hole array region with a threshold of 0.10. When the anomaly score calculated for the corresponding image block exceeds the corresponding threshold, it is determined to be an anomaly, and the three-dimensional coordinate position information of the defect is output.

[0028] In this embodiment, step S6 includes: stacking all the two-dimensional ultrasound images that have completed the above processing according to their spatial positions to form a three-dimensional data set, generating a three-dimensional imaging data image, and obtaining a defect detection result based on the defect features therein; wherein, the defect detection result includes defect attributes, three-dimensional coordinate positions corresponding to the defect attributes, and a complete set of quantization parameters corresponding to the defect attributes. Specifically, the defect attributes include defect type, anomaly score and corresponding anomaly score threshold, and the quantization parameters include the size and morphological features corresponding to the defect.

[0029] Example 3 This embodiment is the third embodiment of a non-destructive testing method for a ceramic matrix composite flame tube. This embodiment is similar to Embodiment 1 or 2, except that it also includes step S7: verifying and calibrating the defect generation model using a standard ceramic matrix composite specimen; wherein, step S7 includes the following steps: S71. Take a standard ceramic matrix composite specimen, which includes pore defects, crack defects, and defects around cooling holes; S72. Use the trained defect generation model to detect standard ceramic matrix composite specimens, generate the resolution and positioning accuracy of the corresponding three-dimensional imaging data image, and obtain the first defect detection result; S73. If the resolution and positioning accuracy obtained in step S72 reach the set threshold, the verification is completed; specifically, the set threshold for resolution is ≥0.15mm, and the set threshold for positioning accuracy is ≤0.07mm. S74. The defects of the standard ceramic matrix composite specimen are calibrated by X-ray to obtain the second defect detection result; the second defect detection result is compared with the first defect detection result obtained in step S72. If the defect identification rate is confirmed to be ≥97%, the calibration is completed.

[0030] Example 4 like Figure 2 The illustration shows an embodiment of a non-destructive testing system according to the present invention, applied to the non-destructive testing method for ceramic matrix composite flame tubes described in any of embodiments one to three. The non-destructive testing system includes a processor 100, a phased array probe 200, a robotic arm 300, and a human-machine interface display screen 400, all communicatively connected to the processor 100. The phased array probe 200 is connected to the drive end of the robotic arm 300. In this embodiment, the processor 100 is selected as an FPGA processor, the phased array probe 200 is selected as a 128-element phased array probe, and the robotic arm 300 is selected as a five-axis linkage robotic arm.

[0031] Specifically, in step S2, the excitation frequency and incident angle of the Lamb wave or surface wave are dynamically adjusted using the adaptive ultrasonic guidance mode of the 128-element phased array probe. The 128-element phased array probe is driven by a five-axis linkage robotic arm to rotate around the flame tube 500 once, thereby acquiring the ultrasonic signal of the flame tube 500. Specifically, in step S3, the acquired ultrasonic signal from the flame tube 500 is amplified and filtered, and then converted into a two-dimensional ultrasonic image set by the FPGA processor; Specifically, in step S4, the FPGA processor processes each two-dimensional ultrasound image in the two-dimensional ultrasound image set to obtain normal samples and abnormal samples. Specifically, in step S5, the FPGA processor constructs a defect generation model based on a convolutional neural network, and normal samples and abnormal samples are input into the defect generation model for training and reconstruction to obtain defect features; Specifically, in step S6, the FPGA processor stacks the two-dimensional ultrasound image set that has undergone the above processing to generate a three-dimensional imaging data image, and obtains the defect detection result based on the defect features therein. The defect detection result can be displayed on the human-computer interaction display screen 400. Specifically, in step S73, if the resolution and positioning accuracy obtained in step S72 do not reach the set threshold, the parameters of the non-destructive testing system are adjusted using a standard ceramic matrix composite specimen; in step S74, if the defect recognition rate is confirmed to be <97%, the parameters of the non-destructive testing system are adjusted using a standard ceramic matrix composite specimen; it should be noted that the parameters of the non-destructive testing system to be adjusted include signal gain, sampling frequency, imaging coordinate offset, and probe scanning step distance parameters.

[0032] In the specific implementation of the above embodiments, the technical features can be combined in any non-contradictory way. For the sake of brevity, not all possible combinations of the above technical features are described. However, as long as the combination of these technical features is not contradictory, it should be considered to be within the scope of this specification.

[0033] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the implementation of the present invention. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively describe all embodiments here. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the claims of the present invention.

Claims

1. A non-destructive testing method for a ceramic matrix composite flame tube, characterized in that, Includes the following steps: S1. Construct a three-dimensional geometric model of the flame tube (500) and a waveguide propagation model of the ceramic matrix composite material; S2. Based on the three-dimensional geometric model and waveguide propagation model, select suitable wave patterns for different regions of the flame tube (500), and dynamically adjust the excitation frequency and incident angle according to the regional characteristics corresponding to the regional geometry to obtain the ultrasonic signal of the flame tube (500). S3. The acquired ultrasonic signal from the flame tube (500) is amplified, filtered, and then converted into a two-dimensional ultrasonic image set; S4. Use a semantic segmentation tool to annotate the pixels of the two-dimensional ultrasound image set to generate a corresponding segmentation mask image; decompose the segmentation mask image into multiple image blocks and classify the image blocks to obtain normal samples and abnormal samples; S5. Construct a defect generation model based on a convolutional neural network, and input the normal samples and abnormal samples into the defect generation model for training and reconstruction to obtain defect features; S6. Stack the two-dimensional ultrasound image set that has undergone the above processing to generate a three-dimensional imaging data image, and obtain the defect detection result based on the defect features therein.

2. The non-destructive testing method for ceramic matrix composite flame tubes according to claim 1, characterized in that, Step S1 includes the following steps: S11. The regional geometry of the flame tube (500) and its corresponding regional characteristics are obtained by laser scanning to construct the three-dimensional geometric model; wherein, the regional geometry of the flame tube (500) includes a straight wall region, a curved transition region, and a cooling hole array region; the regional characteristics include the diameter and wall thickness of the straight wall region, the radius of curvature of the curved transition region, and the diameter and spacing of each cooling hole in the cooling hole array region; S12. Construct the waveguide propagation model based on the acoustic properties of the ceramic matrix composite material; wherein, the acoustic properties of the ceramic matrix composite material include sound velocity, attenuation coefficient, and anisotropy coefficient.

3. The non-destructive testing method for ceramic matrix composite flame tubes according to claim 2, characterized in that, Step S2 includes: for the straight wall region: selecting a Lamb wave and setting a first excitation frequency; for the curved transition region: selecting a Lamb wave and setting a second excitation frequency; for the cooling hole array region: selecting a surface wave and setting a third excitation frequency; wherein the second excitation frequency is greater than the first excitation frequency, and the first excitation frequency is greater than the third excitation frequency; and the incident angle is dynamically adjusted according to the curvature of the region geometry and the sound velocity ratio of the ceramic matrix composite material.

4. The non-destructive testing method for ceramic matrix composite flame tubes according to claim 1, characterized in that, Step S3 includes: using a signal amplifier to gain the acquired ultrasonic signal from the flame tube (500) to a preset multiple, then performing filtering processing through a wavelet transform algorithm, and finally converting it into a two-dimensional ultrasonic image set; wherein, the filtering processing includes: filtering out low-frequency interference through a wavelet transform algorithm and extracting the effective frequency band signal.

5. The non-destructive testing method for ceramic matrix composite flame tubes according to claim 1, characterized in that, Step S4 includes the following steps: S41. The two-dimensional ultrasound image set is pixel-annotated using a semantic segmentation tool. The pixel annotation includes background pixels, defect-free foreground pixels, and defective pixels. Then, a corresponding segmentation mask image is generated. S42. The segmentation mask image is decomposed into multiple image blocks according to a certain pixel size, wherein each image block may contain any one or more pixel labels; S43. Remove some image blocks that contain only background pixels, so that the remaining image blocks have an appropriate ratio of background to defect-free foreground pixels; S44. Classify the remaining image blocks: Image blocks containing defect-free foreground pixels but not defective pixels are classified as normal samples; image blocks containing defective pixels are classified as abnormal samples.

6. The non-destructive testing method for ceramic matrix composite flame tubes according to claim 5, characterized in that, In step S44, for an image block containing defective pixels: The image blocks belonging to the straight wall area and curved transition area of ​​the flame tube (500) are decomposed into multiple image blocks according to a certain pixel size, with the geometric center of the defect pixel as the reference. The image block belonging to the cooling hole array area of ​​the flame tube (500) is expanded to the corresponding hole wall part with the geometric center of the defect pixel as the reference, and then decomposed into multiple image blocks according to a certain pixel size. Then the abnormal samples are obtained.

7. The non-destructive testing method for ceramic matrix composite flame tubes according to claim 1, characterized in that, Step S5 includes the following steps: S51. Construct a defect generation model based on convolutional neural networks; S52. Input the normal samples into the defect generation model, and train it with the goal of minimizing the reconstruction error between the normal samples and the model-generated samples to form generated samples; The normal sample is compared with the generated sample to adjust the reconstruction accuracy of the normal flame tube structure; S53. Input both the normal samples and the abnormal samples into the defect generation model for reconstruction, and calculate the abnormal score using pixel-level mean square error; S54. Set an anomaly score threshold according to the defect type, and determine the corresponding defect type for samples that exceed the corresponding anomaly score threshold.

8. The non-destructive testing method for ceramic matrix composite flame tubes according to claim 7, characterized in that, In step S52, the key indicators of the normal sample and the generated sample are compared to ensure that the reconstruction accuracy of the normal flame tube structure is ≥95%; wherein, the key indicators include: the roundness deviation of the cooling hole array area of ​​the rocket tube <5%, the gray standard deviation of the hole wall portion of the cooling hole array area of ​​the rocket tube <10, and the gray value deviation of the circumferential uniformity of the straight wall area of ​​the rocket tube <10. In step S54, the defect types and their corresponding abnormality score thresholds include: porosity defects with a threshold of 0.06; crack defects with a threshold of 0.08; and defects around cooling holes with a threshold of 0.

10.

9. The non-destructive testing method for a ceramic matrix composite flame tube according to any one of claims 1 to 8, characterized in that, It also includes step S7: verifying and calibrating the defect generation model using a standard ceramic matrix composite specimen; wherein step S7 includes the following steps: S71. Take a standard ceramic matrix composite specimen, wherein the standard ceramic matrix composite specimen includes pore defects, crack defects, and defects around cooling holes; S72. Use the trained defect generation model to detect the standard ceramic matrix composite specimen, generate the resolution and positioning accuracy of the corresponding three-dimensional imaging data image, and obtain the first defect detection result; S73. If the resolution and positioning accuracy obtained in step S72 reach the set threshold, the verification is complete; S74. The defects of the standard ceramic matrix composite specimen are calibrated by X-ray to obtain a second defect detection result; the second defect detection result is compared with the first defect detection result obtained in step S72. If the defect identification rate is confirmed to be ≥97%, the calibration is completed.

10. A system for non-destructive testing of a ceramic matrix composite flame tube according to any one of claims 1 to 9, characterized in that, It includes a processor (100), a phased array probe (200), a robotic arm (300), and a human-machine interaction display screen (400) that are communicatively connected to the processor (100). The phased array probe (200) is connected to the drive end of the robotic arm (300).

Images