Honeycomb sandwich composite structure ultrasonic C scanning imaging and image intelligent classification method

Abstract

The invention relates to an ultrasonic C scanning imaging and image intelligent classification method for a honeycomb sandwich composite structure. The method comprises the following steps: constructing a VGG-16 intelligent classification network; performing ultrasonic C scanning imaging based on the adhesive layer and the honeycomb interface pulse reflection wave; the method comprises the following steps: randomly intercepting a 10 * 10 mm ultrasonic C scanning sample picture, calculating the accumulated sum of the amplitude values of pixel points, and dividing the accumulated sum into three categories of target areas A, B and C; expanding the sample data set by using a data enhancement method, and training the VGG-16 network according to the sample data set; and inputting a to-be-identified C scanning image into the trained VGG-16 network for classification, when the confidence rate of the output target region category reaches a set value, determining that the target region category is the target region category, otherwise, determining that the target region category is an unidentified region. According to the invention, high-precision intelligent identification of the defect ultrasonic C scanning image in the honeycomb sandwich composite structure can be realized, the automation level and accuracy of ultrasonic detection are improved, and the method is widely applicable to composite material detection in the fields of aerospace, ships, automobiles and the like.

Description

Technical Field

[0001] The present invention relates to the technical field of non-destructive testing, and in particular to an ultrasonic C-scanning imaging and image intelligent classification method for a honeycomb sandwich composite structure. Background Art

[0002] Honeycomb sandwich composite materials are widely used in aerospace, automobile, shipbuilding, construction and other fields. Due to their unique light weight and high strength characteristics, honeycomb structures are widely used, especially in the aerospace field. However, due to their complex structure and the involvement of multiple layers of materials, in actual use, honeycomb sandwich composite materials often have problems such as debonding between the adhesive layer and the honeycomb interface and damage to the honeycomb core material. These problems seriously affect the overall performance and safety of the material. Therefore, timely detection and accurate positioning of these defects are of great significance to ensure the safety and reliability of honeycomb sandwich composite materials.

[0003] Traditional detection methods, such as visual inspection and X-ray imaging, can identify defects in honeycomb sandwich composites to a certain extent, but they have some significant limitations. First, traditional methods are not accurate enough in identifying and locating defects, especially in complex honeycomb structures, and often cannot effectively distinguish different types of defects; second, some methods require more complex operating conditions and have blind spots for detecting deep defects inside the material. Most importantly, traditional ultrasonic detection methods usually rely on manual judgment, are inefficient, and are greatly affected by the operator's subjective factors.

[0004] In recent years, with the rapid development of deep learning and artificial intelligence technology, automated detection methods based on image recognition have gradually become an effective solution. Through deep neural networks, especially convolutional neural networks (CNN), deep features can be automatically extracted from ultrasonic C-scan images for defect classification and identification, greatly improving the accuracy and efficiency of detection. At the same time, combined with ultrasonic imaging technology, clearer and more accurate internal defect information of honeycomb sandwich composite materials can be obtained, thereby achieving efficient diagnosis of honeycomb sandwich composite structures.

[0005] Ultrasonic C-scan imaging technology uses the propagation characteristics of ultrasound to obtain detailed information inside honeycomb sandwich composite materials, including the interface between the glue layer and the honeycomb, the state of the honeycomb core, etc. Compared with traditional X-ray or visual inspection methods, ultrasonic testing has stronger penetration and higher safety, and can accurately detect defects inside the material without destroying the material. However, the images produced by ultrasonic imaging technology itself are complex, and the characteristic signals of defects are relatively weak, which makes manual analysis and judgment difficult, especially in the case of multi-layer composite materials, where signal interference is more serious.

[0006] Therefore, developing an efficient and automated method for defect detection of honeycomb sandwich composite structures, and using ultrasonic C-scan imaging combined with deep learning technology for image classification and defect identification, is an effective way to solve current technical problems. Summary of the invention

[0007] In view of the defects of the prior art, the present invention provides a method for ultrasonic C-scan imaging and intelligent image classification of honeycomb sandwich composite structures. It can not only improve the accuracy and efficiency of detection and reduce manual errors, but also accurately classify different types of defects, providing reliable technical support for quality monitoring and safety assessment of honeycomb sandwich composite materials.

[0008] In order to achieve the above invention object, the technical solution adopted by the present invention is as follows:

[0009] A method for ultrasonic C-scanning imaging and intelligent image classification of a honeycomb sandwich composite structure comprises the following steps:

[0010] S1. Build VGG-16 intelligent classification network;

[0011] S2, ultrasonic C-scan imaging based on pulse reflection waves at the interface between the glue layer and the honeycomb;

[0012] S3, randomly intercept the ultrasonic C scan sample image and use the maximum value S of the sum of the pixel amplitude values ​​in the entire square sample image. max The regional categories are divided into benchmarks to obtain the ultrasonic detection data set of honeycomb sandwich composite structure;

[0013] S4, based on the ultrasonic detection dataset of honeycomb sandwich composite structure, train the VGG-16 intelligent classification network;

[0014] S5. Input the re-collected sample image into the trained VGG-16 intelligent classification network. When the confidence rate of the output target area category reaches the set value, it is determined as the output target area category, otherwise it is determined as an unrecognized area.

[0015] Furthermore, in step S1, the VGG-16 intelligent classification network consists of 13 convolutional layers, 3 fully connected layers, 5 pooling layers and a Softmax output layer, which intelligently classifies the C-scan image features of the ultrasonic inspection of the honeycomb sandwich composite structure.

[0016] Furthermore, the honeycomb interface pulse reflection wave in step S2 is extracted by an immersion probe with a center frequency of 15 MHz.

[0017] Furthermore, in step S2, the characteristic signal is tracked to perform ultrasonic C-scan imaging, specifically: the pulse reflection wave of the interface between the glue layer and the honeycomb is tracked through a gate, and ultrasonic C-scan imaging is performed according to the characteristic signal.

[0018] Furthermore, the specific steps of step S3 are:

[0019] S301, randomly intercepting 300 square sample images with a size of 10×10 mm from the ultrasonic C-scan image;

[0020] S302, the maximum value S of the accumulated sum of the amplitude values ​​of the pixels in the entire 10×10 mm square sample image max As a benchmark, three target area categories are divided into:

[0021] The sum of the pixel amplitude values ​​in the square sample image is between S max Between 50% and 60% are classified as target area A;

[0022] The sum of the pixel amplitude values ​​in the square sample image is between S max Between 60% and 75% are classified as target area B;

[0023] The sum of the pixel amplitude values ​​in the square sample image is greater than S max 75% were classified as target area C.

[0024] S303, expanding the sample image into a honeycomb sandwich composite structure ultrasonic detection data set by using one or two data enhancement methods of image rotation, noise injection, and mirroring.

[0025] Furthermore, in step S4, the ultrasonic detection data set of the honeycomb sandwich composite structure is divided into a training set, a validation set and a test set; each data set contains sample images classified as target area A, target area B and target area C, and image intelligent classification network training is performed. The network training parameters are as follows: the number of training times is 150, the initialization learning rate is 0.0003, and the training batch size is 32.

[0026] Furthermore, in step S5, the specific method is:

[0027] S501, performing continuous non-repetitive partitioning operations on the C-scan image to be identified and recording the position coordinates of each partition sample in the C-scan image, where the size of each partition sample is 10×10 mm;

[0028] S502, input the partition sample image into the trained VGG-16 network for intelligent classification task, when the confidence rate of the output target area category reaches 0.6, it is determined as the output target area category, otherwise it is determined as an unrecognized area;

[0029] S503: Reconstruct the image according to the recognition result and position coordinates of the partition sample image. The specific contents of the image reconstruction are as follows:

[0030] The target area is identified as target area A and is marked in green;

[0031] The target area is identified as target area B and is marked in yellow;

[0032] The target area is identified as target area C and marked in red;

[0033] Target areas with recognition confidence lower than 0.6 are marked in white.

[0034] Furthermore, in step S503, the specific content of image reconstruction is as follows: by marking the target area A, the target area B, the target area C and the unidentified area in the image with different colors, an intuitive display of the image intelligent classification result is achieved.

[0035] Furthermore, the target area A, the target area B, and the target area C are divided according to a ratio between the cumulative sum of amplitude values ​​of the sample images and the cumulative sum of maximum amplitude values.

[0036] Compared with the prior art, the advantages of the present invention are:

[0037] 1. The present invention uses VGG-16 convolutional neural network to intelligently classify and extract features from ultrasonic C-scan images, which can effectively distinguish different types of defects in honeycomb sandwich composite materials. Through deep learning algorithms, the network can automatically extract more accurate features from images, avoiding errors and subjective biases that may occur in traditional manual analysis, thereby significantly improving the accuracy of defect detection.

[0038] 2. Traditional honeycomb sandwich composite material detection relies on manual judgment and experience, which has great operational complexity and human errors. The present invention not only realizes automatic identification of defects through intelligent VGG-16 network training and image classification, but also effectively reduces manual intervention and improves the efficiency and consistency of detection. Especially for small defects in complex structures, automated detection can effectively avoid omissions or misjudgments.

[0039] 3. The present invention can quickly process large-scale detection data through characteristic signal tracking and intelligent classification methods based on ultrasonic C-scan images, and divide the defect areas in the ultrasonic image into different target areas (such as target areas A, B, and C), thereby helping to quickly locate and evaluate the severity of the defects. Data enhancement methods (such as image rotation, noise injection, etc.) also greatly increase the diversity of training data sets, improving the generalization ability and recognition accuracy of the model.

[0040] 4. Through the visualization of intelligent classification results, the defect types of different target areas can be intuitively displayed. In the final image, target areas A, B, C and unidentified areas are marked with different colors, which makes it easier for operators to quickly understand and process the inspection results. This image reconstruction method enhances the interpretability and practicality of the results, helping engineers make better decisions.

[0041] 5. The ultrasonic C-scan imaging method used in the present invention can effectively deal with the complexity and multi-layer interface problems of honeycomb sandwich composite structures. By tracking the pulse reflection waves of the adhesive layer and the honeycomb interface, the state of different interfaces can be clearly identified, and defects such as debonding and damage can be accurately located, effectively solving the problems of insufficient penetration and signal interference of traditional ultrasonic imaging in complex structures.

[0042] 6. Compared with the traditional ultrasonic detection method, the intelligent classification network of the present invention can more accurately identify minor defects and effectively avoid false omissions. During the detection process, minor defects are often mistakenly judged as normal or unidentified areas, while the present invention can identify minor defects early, discover potential risks in advance, and improve structural safety.

[0043] 7. Since the present invention adopts an automated classification method based on deep learning, the need for manual analysis is reduced, thereby reducing the time cost and labor intensity of manual operation. At the same time, due to the high efficiency of the model after training, a large number of ultrasonic C-scan images can be processed in a short time, significantly improving the efficiency of detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 It is a flow chart of the ultrasonic C-scan imaging and image intelligent classification method of the honeycomb sandwich composite structure according to an embodiment of the present invention.

[0045] Figure 2 It is a schematic diagram of the VGG-16 network structure of an embodiment of the present invention.

[0046] Figure 3 It is a schematic diagram of water jet type ultrasonic focusing detection according to an embodiment of the present invention.

[0047] Figure 4 The embodiment of the present invention detects the A-scan signal.

[0048] Figure 5 This is the ultrasonic detection C-scan imaging of the embodiment of the present invention.

[0049] Figure 6 1 is an example of sample pictures of three types of target areas in an embodiment of the present invention; (a) is target area A, (b) is target area B, and (c) is target area C;

[0050] Figure 7This is a schematic diagram of the position coordinates of the sample images of the marked partitions according to an embodiment of the present invention;

[0051] Figure 8 is a schematic diagram of relative errors between the intelligent classification method and the mathematical classification method according to an embodiment of the present invention;

[0052] Fig. 9 This is a graph showing the proportion of misidentification types in intelligent classification according to an embodiment of the present invention;

[0053] Fig.10 It is a reconstructed image of the partition sample image recognition result according to an embodiment of the present invention. DETAILED DESCRIPTION

[0054] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings and examples.

[0055] like Figure 1 As shown, a method for ultrasonic C-scanning imaging and intelligent image classification of a honeycomb sandwich composite structure comprises the following steps:

[0056] 1. Construct an intelligent classification network for honeycomb sandwich composite structure images

[0057] like Figure 2 As shown in the figure, a VGG-16 network is constructed to intelligently classify the C-scan image features of ultrasonic testing of honeycomb sandwich composite structures. The VGG-16 network consists of 13 convolutional layers, 3 fully connected layers, 5 pooling layers and a Softmax output layer, where the convolutional and pooling layers are responsible for feature extraction tasks, and the fully connected layers are responsible for classification tasks.

[0058] 2. Extract the characteristic signal of ultrasonic detection of honeycomb sandwich composite structure for C-scan imaging

[0059] The honeycomb sandwich composite structure is subjected to C-scan imaging inspection by the water-jet ultrasonic focusing inspection method. The frequency of the immersion probe is 15MHz, the chip diameter is 6mm, the focal length in water is 19.2mm, and the probe water distance during inspection is 8.56mm. The probe scans back and forth in an "S" shape, extracts an A-scan signal every 0.1mm, and performs color imaging based on the amplitude of the characteristic pulse reflection wave of the A-scan signal. The incident acoustic beam is incident on the water-skin interface, the skin-glue interface, and the glue-honeycomb interface through water coupling; the detection signal is composed of the water-skin interface pulse reflection wave f1, the skin-glue interface pulse reflection wave f2, and the glue layer and honeycomb interface pulse emission wave f3, such as Figure 3 and Figure 4As shown. Based on the pulse reflection wave f3 imaging, the bonding state of the skin and the honeycomb can be displayed. The time interval between the pulse reflection waves f3 and f2 is 0.2μs. When the bonding is intact, the energy of the transmitted wave is strong and the amplitude of the reflected echo is low, which is displayed in blue in the C-scan image. When the bonding is debonded, the energy of the transmitted wave is weak and the amplitude of the reflected echo is high, which is displayed in red in the C-scan image, as shown. Figure 5 (C-scan image) shown.

[0060] 3. Preparation of ultrasonic testing image dataset for honeycomb sandwich composite structures

[0061] A square sample image with a size of 10×10mm was randomly intercepted from the scanned image of the honeycomb sandwich composite structure C, and a total of 300 sample images were intercepted as network training samples. The cumulative sum of the amplitude values ​​of the pixels in the 10×10mm square sample image was calculated, and these square sample images were divided into three types of target areas: target area A, target area B, and target area C. The cumulative sum of the amplitude values ​​of target area A is the smallest, indicating a good bonding area; the cumulative sum of the amplitude values ​​of target area C is the largest, indicating a serious bonding problem area; the cumulative sum of the amplitude values ​​of target area B is between target area A and target area C, indicating a bonding problem area. The classification method of each target area is as follows: the cumulative sum of the amplitude values ​​of the pixels in the 10×10mm square sample image is S; the maximum value of the cumulative sum of the amplitude values ​​of the pixels in the square images in the entire sample is S max ,but:

[0062] (1) The sum of the pixel amplitude values ​​in the square sample image is between S max Between 50% and 60% are classified as target area A;

[0063] (2) The sum of the amplitude values ​​of the pixels in the square sample image is between S max Between 60% and 75% are classified as target area B;

[0064] (3) The sum of the amplitude values ​​of the pixels in the square sample image is greater than S max 75% were classified as target area C.

[0065] The calculation formula for the three types of target areas is as follows:

[0066] Target area A: S max ×50%≤S≤S max ×60% (1)

[0067] Target area B: S max ×60%<S<S max ×75% (2)

[0068] Target area C: S ≥ S max ×75% (3)

[0069] According to equations (1) to (3), among the 300 square sample images, 99 are target area A, 95 are target area B, and 106 are target area C. Then, the data is expanded to 2700 honeycomb sandwich composite structure ultrasonic detection data sets through image rotation, noise injection, mirroring, and the combination of the two methods. Among them, there are 891 sample images divided into target area A, accounting for 33.0%, 855 sample images divided into target area B, accounting for 31.7%, and 954 sample images divided into target area C, accounting for 35.3%.

[0070] Figure 6 The following are example images of sample images of three types of target areas, where (a) is an example image of target area A, (b) is an example image of target area B, and (c) is an example image of target area C.

[0071] 4. Network classification training and testing

[0072] The dataset is randomly divided into training set (train), validation set (val) and test set (test) in a ratio of 7:2:1; each dataset contains sample images classified as target area A, target area B and target area C. The VGG-16 network is trained with the following parameters: 150 training epochs, 0.0003 initial learning rate, and 32 training batch sizes. The VGG-16 network verification accuracy is 94.3%.

[0073] The size of the ultrasonic C-scan image of the honeycomb sandwich composite material to be identified is 180×60 mm. It is divided into 108 10×10 mm partition sample images with continuous non-overlapping partitions, and the position coordinates of each partition sample image in the C-scan image to be identified are recorded to facilitate reassembly after identification (such as Figure 7 These partitioned sample images are input into the trained VGG-16 network, and the target region categories can be output: target region A, target region B, and target region C. When the confidence rate of the output target region category reaches 0.6, it is determined to be the output target region category, otherwise it is determined to be an unrecognized region (i.e., it is impossible to determine the target region category to which it belongs).

[0074] The area and size of each target area obtained by the statistical intelligent classification method are as follows: Figure 8As shown in the figure, and compared with the mathematical classification method, it can be seen that the recognition errors of the intelligent classification method are: 20.0% for target area A, 12.9% for target area B, and 10.8% for target area C. The relative difference of the intelligent classification results of target area A is the largest, and the relative error of the intelligent classification of target area C is the smallest, indicating that the intelligent classification results for the area with serious bonding problems in the honeycomb sandwich composite structure are the best; the proportion of misidentification of each type is counted (such as Fig. 9 As shown in the figure, the misidentification ratios of the target areas A→B and B→C are 4.6% and 5.6% respectively, and the misidentification ratios of the target areas B→A and C→B are both 0.9%. Therefore, the misidentification mainly comes from identifying the mild bonding problem of the honeycomb panel as a severe bonding problem; in addition, the area of ​​the unidentified (uncertain) area accounts for 7.4%. Therefore, although the intelligent recognition method has a larger relative error than the mathematical method, the error comes from identifying mild problems as severe problems, which meets the strict requirements for the evaluation results of non-destructive testing and can effectively avoid missed detections. In order to intuitively display the intelligent classification results, the intelligent classification results of the partitioned sample images are marked as target area A, target area B, target area C, and unidentified areas (confidence rate less than 0.6) in green, yellow, red, and white respectively; and the image expression of the intelligent classification results is obtained by splicing the partitioned sample images according to their original positions, as shown in the figure below. Fig.10 shown.

[0075] The method according to the present invention described above can be implemented in hardware, firmware, or implemented as software or computer code that can be stored in a recording medium (such as a CD ROM, RAM, floppy disk, hard disk or magneto-optical disk), or implemented as a computer code originally stored in a remote recording medium or a non-temporary machine-readable medium downloaded through a network and to be stored in a local recording medium, so that the method described herein can be stored in such software processing on a recording medium using a general-purpose computer, a dedicated processor or programmable or dedicated hardware (such as an ASIC or FPGA). It can be understood that a computer, a processor, a microprocessor controller or programmable hardware includes a storage component (e.g., RAM, ROM, flash memory, etc.) that can store or receive software or computer code. When the software or computer code is accessed and executed by a computer, a processor or hardware, the ultrasonic C-scan imaging and image intelligent classification method of a honeycomb sandwich composite structure described herein is implemented. In addition, when a general-purpose computer accesses the code for implementing the processing shown herein, the execution of the code converts the general-purpose computer into a dedicated computer for executing the processing shown herein.

[0076] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the implementation methods of the present invention, and should be understood that the protection scope of the present invention is not limited to such special statements and embodiments. Those skilled in the art can make various other specific variations and combinations that do not deviate from the essence of the present invention based on the technical revelations disclosed by the present invention, and these variations and combinations are still within the protection scope of the present invention.

Claims

1. A method for ultrasonic C-scan imaging and intelligent image classification of a honeycomb sandwich composite structure, characterized in that: The following steps are involved: S1. Build VGG-16 intelligent classification network; S2, ultrasonic C-scan imaging based on pulse reflection waves at the interface between the glue layer and the honeycomb; S3, randomly intercept the ultrasonic C scan sample image and use the maximum value S of the sum of the pixel amplitude values ​​in the entire square sample image. max The regional categories are divided into benchmarks to obtain the ultrasonic detection data set of honeycomb sandwich composite structure; S4, based on the ultrasonic detection dataset of honeycomb sandwich composite structure, train the VGG-16 intelligent classification network; S5. Input the re-collected sample image into the trained VGG-16 intelligent classification network. When the confidence rate of the output target area category reaches the set value, it is determined as the output target area category, otherwise it is determined as an unrecognized area.

2. The method for ultrasonic C-scan imaging and intelligent image classification of a honeycomb sandwich composite structure according to claim 1, characterized in that: In step S1, the VGG-16 intelligent classification network consists of 13 convolutional layers, 3 fully connected layers, 5 pooling layers and a Softmax output layer, which intelligently classifies the C-scan image features of the ultrasonic inspection of the honeycomb sandwich composite structure.

3. The method for ultrasonic C-scan imaging and intelligent image classification of a honeycomb sandwich composite structure according to claim 1, characterized in that: The honeycomb interface pulse reflection wave in step S2 is extracted by an immersion probe with a center frequency of 15 MHz.

4. The method for ultrasonic C-scan imaging and intelligent image classification of a honeycomb sandwich composite structure according to claim 1, characterized in that: In step S2, the characteristic signal is tracked to perform ultrasonic C-scan imaging, specifically: the pulse reflection wave of the interface between the glue layer and the honeycomb is tracked through a gate, and ultrasonic C-scan imaging is performed according to the characteristic signal.

5. The method for ultrasonic C-scanning imaging and intelligent image classification of a honeycomb sandwich composite structure according to claim 1, characterized in that: The specific steps of step S3 are: S301, randomly intercepting 300 square sample images with a size of 10×10 mm from the ultrasonic C-scan image; S302, taking the maximum value S of the accumulated sum of the amplitude values ​​of the pixels in the entire 10×10 mm square sample image max As a benchmark, three target area categories are divided into: The sum of the pixel amplitude values ​​in the square sample image is between S max Between 50% and 60% are classified as target area A; The sum of the pixel amplitude values ​​in the square sample image is between S max Between 60% and 75% are classified as target area B; The sum of the pixel amplitude values ​​in the square sample image is greater than S max 75% were classified as target area C; S303, expanding the sample image into a honeycomb sandwich composite structure ultrasonic detection data set by using one or two data enhancement methods of image rotation, noise injection, and mirroring.

6. The method for ultrasonic C-scan imaging and intelligent image classification of a honeycomb sandwich composite structure according to claim 1, characterized in that: In step S4, the ultrasonic detection data set of the honeycomb sandwich composite structure is divided into a training set, a validation set and a test set; each data set contains sample images classified as target area A, target area B and target area C, and image intelligent classification network training is performed. The network training parameters are as follows: the number of training times is 150, the initialization learning rate is 0.0003, and the training batch size is 32.

7. The method for ultrasonic C-scanning imaging and intelligent image classification of a honeycomb sandwich composite structure according to claim 1, characterized in that: In step S5, the specific method is: S501, performing continuous non-repetitive partitioning operations on the C-scan image to be identified and recording the position coordinates of each partition sample in the C-scan image, where the size of each partition sample is 10×10 mm; S502, input the partition sample image into the trained VGG-16 network for intelligent classification task, when the confidence rate of the output target area category reaches 0.6, it is determined as the output target area category, otherwise it is determined as an unrecognized area; S503: Reconstruct the image according to the recognition result and position coordinates of the partition sample image. The specific contents of the image reconstruction are as follows: The target area is identified as target area A and is marked in green; The target area is identified as target area B and is marked in yellow; The target area is identified as target area C and marked in red; Target areas with recognition confidence lower than 0.6 are marked in white.

8. The method for ultrasonic C-scan imaging and intelligent image classification of a honeycomb sandwich composite structure according to claim 7, characterized in that: In step S503, the specific content of image reconstruction is as follows: by marking the target area A, the target area B, the target area C and the unidentified area in the image with different colors, an intuitive display of the image intelligent classification result is achieved.

9. The method for ultrasonic C-scan imaging and intelligent image classification of a honeycomb sandwich composite structure according to claim 1, characterized in that: The target area A, the target area B, and the target area C are divided according to the ratio of the accumulated sum of amplitude values ​​of the sample images to the accumulated sum of maximum amplitude values.

Images