Self-adaptive ultrasonic detection and defect identification method for composite material
By using double-sided inspection of composite materials and deep learning technology to automatically identify defects, the problems of blind spots in single-sided inspection and low efficiency of manual identification in composite material inspection have been solved, achieving efficient and accurate defect identification.
Patent Information
- Application Number
- CN202610020951.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-01-08
- Publication Date
- 2026-03-31
AI Technical Summary
Existing methods for detecting defects in composite materials suffer from problems such as blind spots due to single-sided inspection, incomplete detection results, low efficiency of manual judgment, and susceptibility to misjudgment and omission.
An adaptive ultrasonic testing method is used to inspect both sides of the composite material. Deep learning technology is used to automatically determine whether there are defects in the full-focus imaging image. The deep learning model is used for defect classification and localization. The reliability of the inspection is improved by comparing the inspection data of the two sides.
It improves the accuracy and intelligence of defect detection, reduces errors in manual judgment, adapts to large-scale testing needs, and shortens the testing cycle.
Smart Images
Figure CN121762687A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of ultrasonic testing technology, and particularly relates to an adaptive ultrasonic testing and defect identification method for composite materials. Background Technology
[0002] Because composite materials possess excellent properties such as lightweight, high strength, and corrosion resistance, they are widely used in key fields such as aerospace and high-end equipment manufacturing. However, composite materials are prone to various internal defects during molding, processing, and service. These defects can seriously affect the mechanical properties and safety of the materials. Therefore, it is necessary to detect defects in composite materials.
[0003] Currently, when detecting defects in composite materials, adaptive ultrasonic testing and defect identification methods are typically used. Applying ultrasonic testing technology and adaptive correction algorithms to defect identification can effectively overcome the detection interference caused by the anisotropy and layered structure of composite materials by dynamically adjusting detection parameters and correcting propagation errors. This enables precise location, qualitative and quantitative detection and identification of internal defects in materials. It can accurately capture various typical defect signals such as delamination, cracks, pores, and debonding within composite materials. It can not only accurately determine the existence of defects, but also clearly identify key information such as the specific location coordinates, actual size, and geometric features of the defects. This provides comprehensive and accurate data support for subsequent defect repair and performance evaluation, and provides a reliable basis for the quality assessment and safety assurance of composite materials.
[0004] Current ultrasonic testing and defect identification methods typically include the following steps: First, acquiring full matrix data of the composite material. Utilizing the collaborative operation of multiple elements in an ultrasonic phased array device, ultrasonic echo signals within the material are fully acquired through element-by-element excitation and full-element reception, constructing a full matrix data set containing rich defect information. Second, generating elastic constant combinations. Based on the material properties of the composite material, multiple different combinations of elastic constant parameters are preset to provide fundamental parameters for subsequent sound velocity calculations. Third, calculating the original sound velocity curve. Based on the preset elastic constant combinations, the original sound velocity curve of ultrasonic waves propagating within the material is derived and calculated using wave theory. Fourth, obtaining the equivalent sound velocity curve. Combining the propagation environment and boundary conditions during the actual testing process, the original sound velocity curve is corrected and optimized to obtain an equivalent sound velocity curve that better matches the actual propagation law. Then, multiple combinations of elastic constants are evaluated using specific screening criteria to select the optimal elastic constant that best matches the actual testing data. Fifth, based on the optimal elastic constant and the acquired full matrix data, a full-focus imaging algorithm is used to generate a full-focus image of the defect area, completing the entire testing process from data acquisition to image generation.
[0005] The existing ultrasonic testing and defect identification methods have the following problems: 1. Existing methods generally only test one side of the composite material. Due to the layered structure of composite materials and the presence of complex anisotropic properties, single-side testing is prone to producing blind spots, failing to fully cover all areas inside the material. Defects near the untested surface are often difficult to capture accurately, leading to omissions in the test results and failing to fully reflect the true quality of the material. 2. After obtaining a fully focused image of the defect area, existing methods require professional inspectors to manually identify the defect area. This manual identification method not only requires a high level of professional experience and skill from the inspectors, but also has low efficiency and is difficult to adapt to the needs of large-scale, batch testing. At the same time, manual identification is highly subjective and prone to errors due to differences in personal experience, visual fatigue, etc., leading to misjudgment or omission of defects, and has a low level of intelligence. Therefore, it is very necessary to invent an adaptive ultrasonic testing and defect identification method for composite materials. Summary of the Invention
[0006] To address the aforementioned technical problems, this invention provides an adaptive ultrasonic testing and defect identification method for composite materials. This method detects both sides of the composite material and compares the detection data from both sides, improving detection performance and the reliability of defect identification. It is applicable to thicker composite materials, enhances the accuracy of defect detection, and utilizes deep learning technology to determine the presence of defects in the full-focus imaging image, reducing misjudgments or omissions due to manual judgment and improving the level of intelligence.
[0007] An adaptive ultrasonic testing and defect identification method for composite materials specifically includes the following steps:
[0008] Step 1: Preparation before testing;
[0009] Step 2: Collect full matrix data of the unidirectional plate. Using an ultrasonic phased array device, collect full matrix data once along the parallel fiber direction on one side of the unidirectional plate with the same process and material as the composite plate to be tested and with the same fiber direction. Obtain the original signal for calculating the elastic constant and provide reference data for the subsequent sound velocity correction of the composite plate. Only one collection is needed to cover the basic acoustic characteristics of this type of material.
[0010] Step 3: Calculate the core elastic constant of the composite plate to obtain the key parameters describing the anisotropic mechanical properties of the composite plate. Calculating the core elastic constant is the foundation for achieving high-precision imaging, accurate defect quantification and classification, and provides a physical basis for adaptive detection. The core elastic constant of the composite plate includes: longitudinal elastic modulus E. 11 (Along the fiber direction), transverse elastic modulus E 22 / E 33 (Perpendicular to fiber direction), shear modulus G 12 / G13 / G 23 Poisson's ratio 12 / ν 13 / ν 23 ;
[0011] Step 4: Derive the velocity distribution of each ply group in the composite board, establish the mathematical relationship between the elastic constant and the quasi-longitudinal wave group velocity at any propagation angle, and calculate the quasi-longitudinal wave group velocity distribution at different plies and different angles based on the ply sequence and fiber direction of the composite board to be tested. Solve the problem of sound velocity anisotropy in the composite board and provide accurate sound velocity data support for subsequent imaging delay correction.
[0012] Step 5: Collect full matrix data of the composite plate to provide data for subsequent defect imaging;
[0013] Step Six: Adaptive Correction of Imaging Delay. Adaptive correction of imaging delay is performed based on the full matrix data of surface A and surface B of the composite plate. Based on the group velocity distribution obtained in Step Four, for each imaging pixel in the full matrix data of the composite plate, the propagation delay of the ultrasonic wave from emission to focal point to reception is adaptively calculated in combination with its ply position and propagation angle. This corrects the focusing offset caused by the anisotropy and multi-ply structure of the composite plate, ensuring that the delay calculation of each pixel matches the actual ultrasonic propagation law, and obtaining the corrected delay full matrix data of surface A and surface B of the composite plate.
[0014] Step 7: Full-focus imaging and discrimination of defects in composite plates.
[0015] Preferably, in step one, the first step is: the detection system setup and parameter initialization. The setup of the detection system includes a phased array ultrasonic transducer array, an ultrasonic signal acquisition card, a mechanical positioning platform, and a data storage server. The frequency of the phased array ultrasonic transducer array is set to a frequency suitable for the thickness of the composite plate. The sampling rate of the ultrasonic signal acquisition card is not less than 50MHz to ensure signal integrity. The detection probe is arranged along the fiber spindle to ensure stable coupling. An xz coordinate system is established, where x is the scanning direction and z is the thickness direction.
[0016] Step 2: Pre-treat the composite plate sample by removing surface oil and impurities and drying it to avoid interfering with the transmission of ultrasonic signals; mark the detection areas on side A and side B to ensure that the detection areas on the two sides correspond completely.
[0017] Preferably, step three includes:
[0018] Step 1: Extract the echo transit times of quasi-longitudinal and quasi-transverse waves at a 90° propagation angle from the full matrix data of the one-way plate, and calculate the elastic constant C. 22 and C 55 ;
[0019] Step 2: Extract the echo transit time of the quasi-longitudinal wave at a propagation angle of 0° and calculate the elastic constant C. 11 ;
[0020] Step 3: Extract the transit time of the quasi-longitudinal wave between 0° and 90°, and combine it with the obtained constants to solve for the elastic constant C using a global optimization algorithm. 13 .
[0021] Preferably, step five includes:
[0022] Step 1: Collect full matrix data of surface A of the composite panel. Use an ultrasonic phased array device to collect full matrix data of surface A of the composite panel.
[0023] Step 2: Collect full matrix data of the B side of the composite board. Use ultrasonic phased array equipment to collect full matrix data of the B side of the composite board.
[0024] Step 3: Preprocessing for aligning and fusing full matrix data from A and B sides:
[0025] The third step includes:
[0026] S531: Data alignment and pairing: Based on the scanning coordinate information of surfaces A and B, a coordinate mapping algorithm is used to achieve spatial alignment of bidirectional data. That is, with the scanning coordinates of surface A as the reference, the data of surface B is matched according to the corresponding coordinates, and the signals of surface A and surface B at the same spatial position are integrated to form a paired data set.
[0027] S532: Data fusion preprocessing, calculating the cross-correlation coefficient of ultrasound signals from surfaces A and B at the same spatial location;
[0028] S533: Data aggregation. The aligned full matrix data of A and B sides are aggregated to form a full matrix dataset covering the full thickness of the composite plate, which is used for subsequent adaptive correction imaging delay steps.
[0029] Preferably, step seven includes:
[0030] Step 1: Generate a full-focus imaging image. Apply the full-matrix data of the correction delay on side A of the composite plate and the full-matrix data of the correction delay on side B of the composite plate to perform signal delay superposition processing, and generate imaging images of side A and side B of the composite plate respectively. The imaging images clearly show the location, size and shape of defects such as delamination, cracks and pores inside the composite plate, and achieve accurate characterization of defects.
[0031] Step 2: Defect detection. After obtaining the full-focus image, deep learning technology is used to automatically determine whether the full-focus image contains defects.
[0032] The second step includes:
[0033] S721: Construction of training dataset for deep learning model: Collect full-focus images of the A and B sides of different types of composite materials, covering defect-free samples and typical defect samples, with ≥500 images of each type of defect, and the image resolution of the samples is set to a uniform value;
[0034] Defective samples were labeled using image annotation tools;
[0035] We pre-trained the defect-free samples using an autoencoder to learn the feature distribution of normal images and generate synthetic samples that conform to normal features. We used a generative adversarial network to generate synthetic defect samples of different types and sizes to supplement the lack of real defect samples. We divided the expanded dataset into training, validation and test sets according to the proportion. The training set is used to train model parameters, the validation set is used to adjust model hyperparameters, and the test set is used to evaluate model performance.
[0036] S722: Defect detection: Using a deep learning model architecture for image tasks, establish a baseline and determine whether defects exist;
[0037] S723: Defect localization and classification: Using a visual model tool, bounding boxes are marked to achieve localization and preliminary classification. The images of the defective A-side and B-side are input into the trained visual model tool. The model automatically outputs the bounding box, defect type and confidence level of each defect. The bounding boxes are marked on the image.
[0038] S724: Pixel-level segmentation: Pixel-level segmentation is performed using a fully convolutional neural network architecture. The defect area output by the visual model tool is cropped and then input into the fully convolutional neural network architecture model to obtain a pixel-level segmentation map. The quantitative parameters of the defect are calculated through the segmentation map: area, size, and morphological parameters.
[0039] S725: Improve the generalization of the model by continuously generating new defect samples through generative adversarial networks to expand the defect type coverage of the dataset; at the same time, optimize the model by transfer learning, transferring the pre-trained model parameters to the composite material detection task to reduce the time cost of retraining deep learning models.
[0040] S726: Output defect detection results: Compare and analyze the defect detection results of surface A and surface B; output a defect detection report, including the location coordinates, type, detection confidence level and other information of the defect, and generate an image of the marked defect area.
[0041] Beneficial effects
[0042] Compared with the prior art, the present invention has the following beneficial effects:
[0043] 1. By inspecting the A and B sides of the composite material, full-focus images are obtained respectively. Deep learning technology can be used to automatically determine whether the full-focus images of both sides contain defects, improving the automation and intelligence level of defect identification in full-focus images. The deep learning model can automatically discover subtle defect features in full-focus images that are difficult for the human eye to identify, reducing the reliance on traditional manual judgment and effectively avoiding missed or false detections caused by subjective judgment bias. By training on a large amount of composite material defect imaging data of different types and sizes, the deep learning model can achieve rapid and accurate classification and localization of various defects, significantly shortening the detection cycle and adapting to the real-time requirements of adaptive ultrasonic testing. When faced with imaging interference or new defect types under complex working conditions, it can continuously optimize the discrimination performance through incremental learning, providing reliable technical support for the efficiency and accuracy of ultrasonic testing of composite materials.
[0044] 2. By inspecting both the A and B sides of the composite material and comparing the inspection data from both sides, the inspection performance and the reliability of defect identification are improved. Attached Figure Description
[0045] Figure 1 This is a flowchart of an adaptive ultrasonic testing and defect identification method for composite materials.
[0046] Figure 2 This is a flowchart of the preparations before testing.
[0047] Figure 3 This is a flowchart for calculating the core elastic constant of a composite plate.
[0048] Figure 4 This is a flowchart for collecting full matrix data from the composite board.
[0049] Figure 5 This is a flowchart of full-focus imaging and discrimination of defects in composite plates. Detailed Implementation
[0050] The present invention will be further described below with reference to the accompanying drawings:
[0051] In the picture:
[0052] As attached Figure 1 and attached Figure 2 As shown:
[0053] An adaptive ultrasonic testing and defect identification method for composite materials specifically includes the following steps:
[0054] S1: Preparations before testing;
[0055] S11: Detection system setup and parameter initialization. The detection system setup includes a phased array ultrasonic transducer array, an ultrasonic signal acquisition card, a mechanical positioning platform, and a data storage server. The frequency of the phased array ultrasonic transducer array is set to a frequency suitable for the thickness of the composite plate. The sampling rate of the ultrasonic signal acquisition card is not less than 50MHz to ensure signal integrity. The detection probe is arranged along the fiber spindle to ensure stable coupling. An xz coordinate system is established, where x is the scanning direction and z is the thickness direction.
[0056] S12: Pre-treat the composite plate sample to remove surface oil and impurities and dry it to avoid interfering with the transmission of ultrasonic signals; mark the detection areas on side A and side B to ensure that the detection areas on the two sides are completely corresponding.
[0057] S2: Collect full matrix data of the unidirectional plate. Using an ultrasonic phased array device, collect full matrix data once along the parallel fiber direction on one side of the unidirectional plate with the same process and material as the composite plate to be tested and with the same fiber direction. Obtain the original signal for calculating the elastic constant and provide reference data for the sound velocity correction of the subsequent composite plate. Only one collection is needed to cover the basic acoustic characteristics of this type of material.
[0058] S3: Calculate the core elastic constant of the composite plate to obtain key parameters describing its anisotropic mechanical properties. Calculating the core elastic constant is fundamental for achieving high-precision imaging, accurate defect quantification and classification, and provides a physical basis for adaptive detection. The core elastic constant of the composite plate includes: longitudinal elastic modulus E. 11 (Along the fiber direction), transverse elastic modulus E 22 / E 33 (Perpendicular to fiber direction), shear modulus G 12 / G 13 / G 23 Poisson's ratio 12 / ν 13 / ν 23 ;
[0059] S4: Derive the velocity distribution of each ply group in the composite board, establish the mathematical relationship between the elastic constant and the quasi-longitudinal wave group velocity at any propagation angle, calculate the quasi-longitudinal wave group velocity distribution of different plies and different angles according to the ply sequence and fiber direction of the composite board to be tested, solve the sound velocity anisotropy problem of the composite board, and provide accurate sound velocity data support for subsequent imaging delay correction.
[0060] S5: Collect full matrix data of the composite plate to provide data for subsequent defect imaging;
[0061] S6: Adaptive correction of imaging delay. Based on the full matrix data of the composite plate A side and the full matrix data of the composite plate B side, adaptive correction of imaging delay is performed. Based on the group velocity distribution obtained in S4, for each imaging pixel in the full matrix data of the composite plate, combined with its ply position and propagation angle, the propagation delay of ultrasonic wave transmission-focusing point-reception is adaptively calculated to correct the focusing offset caused by the anisotropy and multi-ply structure of the composite plate, ensuring that the delay calculation of each pixel matches the actual ultrasonic propagation law, and obtaining the full matrix data of correction delay of the composite plate A side and the full matrix data of correction delay of the composite plate B side.
[0062] S7: Full-focus imaging and discrimination of defects in composite plates.
[0063] In this implementation plan, in conjunction with the appendix Figure 3 As shown, S3 includes:
[0064] S31: Extract the echo transit times of quasi-longitudinal and quasi-transverse waves at a 90° propagation angle from the full matrix data of the one-way plate, and calculate the elastic constant C. 22 and C 55 ;
[0065] S32: Extract the echo transit time of the quasi-longitudinal wave at a propagation angle of 0° and calculate the elastic constant C. 11 ;
[0066] S33: Extract the transit time of the quasi-longitudinal wave between 0° and 90°, and combine it with the obtained constants to solve for the elastic constant C using a global optimization algorithm. 13 .
[0067] In this implementation plan, in conjunction with the appendix Figure 4 As shown, S5 includes:
[0068] S51: Acquire full matrix data of surface A of the composite plate. Using an ultrasonic phased array device, acquire full matrix data of surface A of the composite plate. The phased array transducer array is attached to the marked area of surface A, and a one-to-one full matrix acquisition mode is used for acquisition. A single array element in the transducer array is controlled as the transmitting element, and all other array elements are sequentially used as the receiving elements to complete a set of signal acquisition. The transmitting elements are replaced one by one until all array elements have completed the transmission action, forming a full matrix data set of surface A. During the acquisition process, the transducer array is moved along a preset path by a mechanical positioning platform to cover the entire detection area of surface A. The coordinate information of each scanning position and the corresponding full matrix data are recorded synchronously. The data format is stored in binary format, including the transmitting element number, receiving element number, scanning coordinates, and signal time sequence information. After the acquisition is completed, the data is pre-processed: bandpass filtering and signal amplitude normalization are used to remove invalid signals.
[0069] S52: Acquire full matrix data of the composite plate's B side. Using an ultrasonic phased array device, acquire full matrix data of the composite plate's B side. Keeping the composite plate stationary, transfer the phased array transducer array to the B side. Through laser positioning calibration, ensure that the scanning area of the transducer array is aligned with the A side, with a coordinate deviation ≤0.1mm. Using the same basic parameters acquired from the A side, adopt the same one-transmit, one-receive full matrix acquisition mode and scanning path as the A side to complete the acquisition of full matrix data of the B side. During the acquisition process, synchronously record the coordinate information and full matrix data of each scanning position on the B side. Perform preliminary preprocessing such as bandpass filtering, amplitude normalization, and invalid signal removal. Monitor the signal-to-noise ratio of the ultrasonic signal in real time during the acquisition process. If the signal-to-noise ratio is lower than the preset threshold, adjust the signal gain or re-clean the contact surface to ensure that the data quality of the A and B sides is consistent.
[0070] S53: Preprocessing for alignment and fusion of full matrix data from A and B sides:
[0071] S531: Data alignment and pairing: Based on the scanning coordinate information of surfaces A and B, a coordinate mapping algorithm is used to achieve spatial alignment of bidirectional data. That is, with the scanning coordinates of surface A as the reference, the data of surface B is matched according to the corresponding coordinates, and the signals of surface A and surface B at the same spatial position are integrated to form a paired data set.
[0072] S532: Data fusion preprocessing, calculates the cross-correlation coefficient of ultrasonic signals from surfaces A and B at the same spatial location, removes abnormal data with a cross-correlation coefficient below 0.8 to avoid signal deviation caused by unstable contact surfaces; performs signal superposition and averaging on effective paired data to enhance the target signal intensity and suppress random noise;
[0073] S533: Data aggregation. The aligned full matrix data of A and B sides are aggregated to form a full matrix dataset covering the full thickness of the composite plate, which is used for subsequent adaptive correction imaging delay steps.
[0074] In this implementation plan, in conjunction with the appendix Figure 5 As shown, S7 includes:
[0075] S71: Generate a full-focus imaging image. Apply the full-focus algorithm to the full-matrix data of the correction delay of the composite plate A side and the full-matrix data of the correction delay of the composite plate B side to perform signal delay superposition processing, and generate imaging images of the composite plate A side and the composite plate B side respectively. The imaging images clearly show the location, size and shape of defects such as delamination, cracks and pores inside the composite plate, and realize the accurate characterization of defects.
[0076] S72: Defect detection. After obtaining the full-focus image, deep learning technology is used to automatically determine whether the full-focus image contains defects.
[0077] S721: Construction of training dataset for deep learning model: Collect full-focus images of the A and B sides of different types of composite materials, covering defect-free samples and typical defect samples, with ≥500 images of each type of defect, and the image resolution of the samples is set to a uniform value;
[0078] Defect samples are labeled using an image annotation tool, such as the LabelImg tool. For classification tasks, labels are assigned to indicate whether the samples are defect-free or defective. For localization tasks, the bounding box coordinates and defect type of the defect region are annotated. For segmentation tasks, pixel-level annotation is used to define the pixel range of the defect region.
[0079] We pre-trained the defect-free samples using an autoencoder to learn the feature distribution of normal images and generate synthetic samples that conform to normal features. We used a generative adversarial network to generate synthetic defect samples of different types and sizes to supplement the lack of real defect samples. We divided the expanded dataset into training, validation and test sets according to the proportion. The training set is used to train model parameters, the validation set is used to adjust model hyperparameters, and the test set is used to evaluate model performance.
[0080] S722: Defect Judgment: Using a deep learning model architecture for image tasks, a baseline is established to determine whether defects exist. Full-focus images of surface A and surface B are sequentially input into the baseline model, which outputs the probability of defect existence for each image. Images with defects are filtered out and proceed to the subsequent defect localization and classification steps. Images without defects are directly marked as qualified. The specific deep learning model architecture used for image tasks can be ResNet50 / EfficientNet-B0.
[0081] S723: Defect Localization and Classification: Using a visual model tool, bounding boxes are labeled to achieve localization and preliminary classification. The images of the defective A-side and B-side are input into the trained visual model tool, and the model automatically outputs the bounding box, defect type, and confidence level for each defect. The bounding boxes are marked on the images. For high-confidence defects with a confidence level ≥ 0.9, their location and type are directly recorded. For suspected defects with a confidence level between 0.7 and 0.9, the bounding box information is retained, and the defect is further verified in subsequent pixel-level segmentation steps. YOLOv8 can be used as the specific visual model tool.
[0082] S724: Pixel-level segmentation: Pixel-level segmentation is performed using a fully convolutional neural network architecture. The defect region output by the visual model tool is cropped and then input into the fully convolutional neural network architecture to obtain a pixel-level segmentation map. Quantitative parameters of the defect are calculated from the segmentation map: area, size, and morphological parameters. For suspected defects, if the area of the segmented defect region is greater than or equal to a preset threshold, it is determined to be a real defect, and its quantitative parameters are recorded. Otherwise, it is determined to be a false detection and is removed. The U-Net model can be used as the specific fully convolutional neural network architecture.
[0083] S725: Improves model generalization by continuously generating new defect samples through generative adversarial networks, expanding the defect type coverage of the dataset; at the same time, it optimizes the model through transfer learning, transferring the pre-trained model parameters to the composite material detection task, reducing the time cost of retraining. The deep learning model significantly shortens the detection cycle, adapts to the real-time requirements of adaptive ultrasonic testing, and can continuously optimize the discrimination performance through incremental learning when facing imaging interference or new defect types under complex working conditions, providing reliable technical support for the efficiency and accuracy of ultrasonic testing of composite materials.
[0084] S726: Output Defect Detection Results: Compare and analyze the defect detection results of surface A and surface B. Based on spatial coordinate alignment information, determine whether it is the projection of the same defect onto the imaging images of both surfaces. For the same defect, fuse the quantitative parameters of both surfaces, taking the average of area and size to improve quantitative accuracy. For defects detected only on one surface, combine the composite plate thickness and ultrasonic propagation path to determine whether it is a surface defect or an internal defect. For example, if it is detected only on surface A and close to the surface, it is determined to be a surface defect on surface A. Output a defect detection report, including the defect's location coordinates, type, detection confidence level, etc., and generate an imaging image of the marked defect area to provide an intuitive basis for subsequent processing.
[0085] Any technical solution that achieves the above-mentioned technical effects by utilizing the technical solutions described in this invention, or by designing similar technical solutions by those skilled in the art under the inspiration of the technical solutions described in this invention, falls within the protection scope of this invention.
Claims
1. An adaptive ultrasonic testing and defect identification method for composite materials, characterized in that, The adaptive ultrasonic testing and defect identification method for this composite material specifically includes the following steps: Step 1: Preparation before testing; Step 2: Collect full matrix data of the unidirectional plate. Using an ultrasonic phased array device, collect full matrix data once along the parallel fiber direction on one side of the unidirectional plate with the same process and material as the composite plate to be tested and with the same fiber direction. Obtain the original signal for calculating the elastic constant and provide reference data for the subsequent sound velocity correction of the composite plate. Only one collection is needed to cover the basic acoustic characteristics of this type of material. Step 3: Calculate the core elastic constant of the composite plate to obtain the key parameters describing the anisotropic mechanical properties of the composite plate. Calculating the core elastic constant is the foundation for achieving high-precision imaging, accurate defect quantification and classification, and provides a physical basis for adaptive detection. The core elastic constant of the composite plate includes: longitudinal elastic modulus E. 11 (Along the fiber direction), transverse elastic modulus E 22 / E 33 (Perpendicular to fiber direction), shear modulus G 12 / G 13 / G 23 Poisson's ratio 12 / ν 13 / ν 23 ; Step 4: Derive the velocity distribution of each ply group in the composite board, establish the mathematical relationship between the elastic constant and the quasi-longitudinal wave group velocity at any propagation angle, and calculate the quasi-longitudinal wave group velocity distribution at different plies and different angles based on the ply sequence and fiber direction of the composite board to be tested. Solve the problem of sound velocity anisotropy in the composite board and provide accurate sound velocity data support for subsequent imaging delay correction. Step 5: Collect full matrix data of the composite plate to provide data for subsequent defect imaging; Step Six: Adaptive Correction of Imaging Delay. Adaptive correction of imaging delay is performed based on the full matrix data of surface A and surface B of the composite plate. Based on the group velocity distribution obtained in Step Four, for each imaging pixel in the full matrix data of the composite plate, the propagation delay of the ultrasonic wave from emission to focal point to reception is adaptively calculated in combination with its ply position and propagation angle. This corrects the focusing offset caused by the anisotropy and multi-ply structure of the composite plate, ensuring that the delay calculation of each pixel matches the actual ultrasonic propagation law, and obtaining the corrected delay full matrix data of surface A and surface B of the composite plate. Step 7: Full-focus imaging and discrimination of defects in composite plates.
2. The adaptive ultrasonic testing and defect identification method for composite materials as described in claim 1, characterized in that, In step one, the first step is: the detection system setup and parameter initialization. The setup of the detection system includes a phased array ultrasonic transducer array, an ultrasonic signal acquisition card, a mechanical positioning platform, and a data storage server. The frequency of the phased array ultrasonic transducer array is set to a frequency suitable for the thickness of the composite plate. The sampling rate of the ultrasonic signal acquisition card is not less than 50MHz to ensure signal integrity. The detection probe is arranged along the fiber spindle to ensure stable coupling. An xz coordinate system is established, where x is the scanning direction and z is the thickness direction. Step 2: Pre-treat the composite plate sample by removing surface oil and impurities and drying it to avoid interfering with the transmission of ultrasonic signals; mark the detection areas on side A and side B to ensure that the detection areas on the two sides correspond completely.
3. The adaptive ultrasonic testing and defect identification method for composite materials as described in claim 1, characterized in that, Step three includes: Step 1: Extract the echo transit times of quasi-longitudinal and quasi-transverse waves at a 90° propagation angle from the full matrix data of the one-way plate, and calculate the elastic constant C. 22 and C 55 ; Step 2: Extract the echo transit time of the quasi-longitudinal wave at a propagation angle of 0° and calculate the elastic constant C. 11 ; Step 3: Extract the transit time of the quasi-longitudinal wave between 0° and 90°, and combine it with the obtained constants to solve for the elastic constant C using a global optimization algorithm. 13 .
4. The adaptive ultrasonic testing and defect identification method for composite materials as described in claim 1, characterized in that, Step five includes: Step 1: Collect full matrix data of surface A of the composite panel. Use an ultrasonic phased array device to collect full matrix data of surface A of the composite panel. Step 2: Collect full matrix data of the B side of the composite board. Use ultrasonic phased array equipment to collect full matrix data of the B side of the composite board. Step 3: Preprocessing for aligning and fusing full matrix data from A and B sides: The third step includes: S531: Data alignment and pairing: Based on the scanning coordinate information of surfaces A and B, a coordinate mapping algorithm is used to achieve spatial alignment of bidirectional data. That is, with the scanning coordinates of surface A as the reference, the data of surface B is matched according to the corresponding coordinates, and the signals of surface A and surface B at the same spatial position are integrated to form a paired data set. S532: Data fusion preprocessing, calculating the cross-correlation coefficient of ultrasound signals from surfaces A and B at the same spatial location; S533: Data aggregation. The aligned full matrix data of A and B sides are aggregated to form a full matrix dataset covering the full thickness of the composite plate, which is used for subsequent adaptive correction imaging delay steps.
5. The adaptive ultrasonic testing and defect identification method for composite materials as described in claim 1, characterized in that, Step seven includes: Step 1: Generate a full-focus imaging image. Apply the full-matrix data of the correction delay on side A of the composite plate and the full-matrix data of the correction delay on side B of the composite plate to perform signal delay superposition processing, and generate imaging images of side A and side B of the composite plate respectively. The imaging images clearly show the location, size and shape of defects such as delamination, cracks and pores inside the composite plate, and achieve accurate characterization of defects. Step 2: Defect detection. After obtaining the full-focus image, deep learning technology is used to automatically determine whether the full-focus image contains defects. The second step includes: S721: Construction of training dataset for deep learning model: Collect full-focus images of the A and B sides of different types of composite materials, covering defect-free samples and typical defect samples, with ≥500 images of each type of defect, and the image resolution of the samples is set to a uniform value; Defective samples were labeled using image annotation tools; We pre-trained the defect-free samples using an autoencoder to learn the feature distribution of normal images and generate synthetic samples that conform to normal features. We used a generative adversarial network to generate synthetic defect samples of different types and sizes to supplement the lack of real defect samples. We divided the expanded dataset into training, validation and test sets according to the proportion. The training set is used to train model parameters, the validation set is used to adjust model hyperparameters, and the test set is used to evaluate model performance. S722: Defect detection: Using a deep learning model architecture for image tasks, establish a baseline and determine whether defects exist; S723: Defect localization and classification: Using a visual model tool, bounding boxes are marked to achieve localization and preliminary classification. The images of the defective A-side and B-side are input into the trained visual model tool, and the model automatically outputs the bounding box, defect type and confidence level of each defect. The bounding boxes are marked on the image. S724: Pixel-level segmentation: Using a fully convolutional neural network architecture, the defect area output by the visual model tool is cropped and then input into the fully convolutional neural network architecture to obtain a pixel-level segmentation map. The quantitative parameters of the defect are calculated through the segmentation map: area, size, and morphological parameters. S725: Improve the generalization of the model by continuously generating new defect samples through generative adversarial networks to expand the defect type coverage of the dataset; at the same time, optimize the model by transfer learning, transferring the pre-trained model parameters to the composite material detection task to reduce the time cost of retraining deep learning models. S726: Output defect detection results: Compare and analyze the defect detection results of surface A and surface B; output a defect detection report, including the location coordinates, type, detection confidence level and other information of the defect, and generate an image of the marked defect area.