Quantitative Analysis Method for Internal Defects of Carbon Fiber Honeycomb Sandwich Based on Artificial Intelligence

Abstract

The present invention belongs to the technical field of nondestructive testing and identification, and discloses a method for quantitative analysis of internal defects of a carbon fiber honeycomb sandwich based on artificial intelligence, comprising the following steps: obtaining an infrared nondestructive testing image; using a YOLO neural network to perform defect image segmentation to obtain the pixel value of each pixel of the mask; calibrating the mask threshold by setting the mask threshold to control the generated size of the mask and obtaining the calibrated mask threshold; obtaining an infrared nondestructive testing image of a workpiece to be detected under the same detection conditions, inputting it into the neural network and applying the calibrated mask threshold to obtain an accurate defect boundary; calculating the accurate defect area by multiplying the number of mask pixels by the actual area of the pixels, so as to realize the quantitative analysis of the internal defects of the workpiece. The present invention adopts the above method for quantitative analysis of internal defects of a carbon fiber honeycomb sandwich based on artificial intelligence, realizes rapid large-batch detection and quantitative analysis of defects, and provides technical support for the quality control of composite materials.

Description

Technical Field

[0001] The present invention relates to the technical field of non-destructive testing and identification, and particularly to a method for quantitatively analyzing internal defects of carbon fiber honeycomb sandwich based on artificial intelligence. Background Art

[0002] With the continuous improvement of the requirements for material performance in modern industry, carbon fiber honeycomb sandwich composites are widely used in the fields of aerospace, high-speed trains, automotive manufacturing, etc. due to their light weight, high strength, excellent rigidity and impact resistance. These materials usually consist of thin and strong carbon fiber panels and lightweight honeycomb cores. This structure not only provides high strength and stiffness, but also reduces the structural weight and improves energy efficiency. However, internal defects such as delamination and holes are likely to occur during the manufacturing process of carbon fiber honeycomb sandwich composites. These defects seriously affect the mechanical properties and structural integrity of the materials, increasing safety risks. Traditional detection methods, such as ultrasonic testing and X-ray testing, have problems such as slow detection speed, high cost, complex operation, or potential hazards to the environment and personnel, and their accuracy and efficiency in quantitatively analyzing defects need to be improved.

[0003] Infrared non-destructive testing technology uses the principle of infrared thermal imaging to infer internal defects by detecting thermal anomalies on the surface of materials. This method has the advantages of being fast and non-contact, and is particularly suitable for materials that are sensitive to the environment or difficult to detect by traditional methods. However, existing infrared non-destructive testing technologies still face challenges in defect detection and quantitative analysis of carbon fiber honeycomb sandwich composites, including how to accurately identify the boundaries of defects and how to achieve precise measurement of defect size and area.

[0004] With the development of artificial intelligence technology, image recognition and analysis technology based on deep learning provides new possibilities for defect detection of composite materials. As an efficient object detection algorithm, the YOLO neural network has received attention in the field of industrial detection due to its fast speed and high accuracy. Although defects in infrared images can be segmented using YOLO, since the boundaries of defects in infrared images are relatively blurred, the defect boundaries segmented by artificial intelligence are often not accurate enough, resulting in problems such as the inability to accurately measure defect size and area. Summary of the Invention

[0005] The purpose of the present invention is to provide a method for quantitatively analyzing internal defects of carbon fiber honeycomb sandwich based on artificial intelligence, to solve the problem of inaccurate defect boundary recognition in infrared images by artificial intelligence technology, and to provide technical support for the quality control of composite materials.

[0006] To achieve the above purpose, the present invention provides a method for quantitatively analyzing internal defects of carbon fiber honeycomb sandwich based on artificial intelligence, including the following steps:

[0007] Step S1: Obtain the infrared non-destructive testing image of a standard specimen of carbon fiber sandwich composite with a known defect area;

[0008] Step S2: Use the YOLO neural network to perform defect image segmentation on the infrared non-destructive testing image obtained in Step S1, and obtain the value of each pixel point of the mask in the image segmentation result;

[0009] Step S3: Control the generated size of the mask by setting the mask threshold, and calibrate the mask threshold using the defect with a known area to obtain the calibrated mask threshold;

[0010] Step S4: Under the same infrared non-destructive testing conditions, obtain the infrared non-destructive testing image of the workpiece to be detected, input it into the neural network and apply the calibrated mask threshold to obtain an accurate defect boundary;

[0011] Step S5: Calculate the number of mask pixel points and multiply it by the actual area of the pixel points to obtain an accurate defect area, thereby realizing the quantitative analysis of the internal defects of the carbon fiber honeycomb sandwich composite.

[0012] Preferably, in Step S1, use an infrared non-destructive testing device to detect the standard specimen with a known defect area and obtain the infrared non-destructive testing image;

[0013] When obtaining the infrared non-destructive testing image, the infrared device should be directly facing the standard specimen for detection to ensure that the obtained workpiece and defect images do not distort and deform, and avoid changes in the shape and proportion of the workpiece and defect images.

[0014] Preferably, the infrared non-destructive testing system includes an infrared thermal excitation module, an infrared thermal imaging acquisition module, and a control and post-processing module.

[0015] Preferably, in Step S2, use the YOLO neural network to perform defect image segmentation on the infrared non-destructive testing image obtained in Step S1, and obtain the value of each pixel point of the mask in the image segmentation result. The specific process is as follows:

[0016] Step S21: First, make a dataset of infrared non-destructive defects. The defects in the dataset are framed and marked with polygons, and the polygons are larger than the defect boundaries;

[0017] Step S22: Then, input the dataset into the YOLO image segmentation neural network for training and obtain the trained neural network;

[0018] Step S23: Finally, use the trained neural network to perform image segmentation on the infrared non-destructive testing image of the standard specimen obtained in Step S1. The image segmentation result will include the mask of the defect;

[0019] Among them, the essence of the mask is a matrix, and the matrix contains the values of each mask pixel point.

[0020] Preferably, in step S3, the generation size of the mask is controlled by setting a mask threshold, and the mask threshold is calibrated using a defect with a known area to obtain a calibrated mask threshold. The specific process is as follows:

[0021] Step S31: Based on the matrix of the defect mask obtained in step S2, set a mask threshold. When the value in the matrix is less than the set defect threshold, remove the mask pixel points less than the threshold to control the generation size of the final mask.

[0022] Step S32: Calibrate the mask threshold using a defect with a known area to obtain a calibrated mask threshold.

[0023] After setting a certain mask threshold, count the number of mask pixel points and multiply it by the actual area of the pixel point to obtain the defect area under this threshold. After traversing the mask thresholds within the range, obtain the defect area under each threshold, and the threshold with the smallest error from the actual defect area is the calibrated mask threshold.

[0024] Preferably, in step S4, under the same infrared non-destructive testing conditions, obtain the infrared non-destructive testing image of the workpiece to be detected, input it into the neural network and apply the calibrated mask threshold to obtain an accurate defect boundary. The specific process is as follows:

[0025] Step S41: Under the same infrared non-destructive testing conditions as in step S1, obtain the infrared non-destructive testing image of the workpiece to be detected.

[0026] Among them, the detection parameters include detection distance and angle, ambient temperature, acquisition frequency, exposure time, focal length, field of view, filtering parameters, and imaging parameters.

[0027] Step S42: Input the obtained infrared non-destructive testing image into the neural network for prediction and apply the calibrated mask threshold. If there is a defect, obtain an accurate defect segmentation mask.

[0028] Preferably, in step S5, according to the mask of the defect obtained in step S4, calculate the number of mask pixel points and multiply it by the actual area of the pixel point to obtain an accurate defect area, thereby realizing the quantitative analysis of the internal defects of the carbon fiber honeycomb sandwich composite material.

[0029] Therefore, the present invention adopts the above-mentioned method for quantitative analysis of internal defects of carbon fiber honeycomb sandwich based on artificial intelligence, and the beneficial effects are as follows:

[0030] (1) The present invention utilizes infrared non-destructive testing technology combined with artificial intelligence algorithms. Through the YOLO neural network, defect image segmentation is performed on infrared images. After using the calibration method of the mask threshold, more accurate quantitative analysis of defects can be carried out.

[0031] (2) The present invention uses artificial intelligence methods. Only by setting the calibrated threshold, under the same detection conditions, rapid and large-scale detection and defect quantitative analysis tasks can be achieved, providing technical support for the quality control of composite materials.

[0032] The technical solution of the present invention will be further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 is a flowchart of the method for quantitatively analyzing internal defects of a carbon fiber honeycomb sandwich based on artificial intelligence according to the present invention;

[0034] Figure 2 is a calibration curve of the mask threshold according to the present invention;

[0035] Figure 3 is for the present invention original infrared non-destructive test images of defects and masks under different mask thresholds of the present invention; wherein, (a) is the original image; (b) is the mask with a mask threshold of 0.1; (c) is the mask with a mask threshold of 0.66; (d) is the mask with a mask threshold of 0.9. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0036] The technical solution of the present invention will be further described below through the accompanying drawings and embodiments.

[0037] As Figure 1 shown, the method for quantitatively analyzing internal defects of a carbon fiber honeycomb sandwich based on artificial intelligence according to the present invention includes the following steps:

[0038] Step S1, obtain an infrared non-destructive test image of a standard test block of carbon fiber sandwich composite material with a known defect area.

[0039] Use an infrared non-destructive testing device to detect the standard test block with a known defect area and obtain an infrared non-destructive test image. When obtaining the infrared non-destructive test image, the infrared device should be directly facing the standard test block for detection to ensure that the obtained workpiece and defect images are not distorted and deformed as much as possible, and to avoid changes in the shape and proportion of the workpiece and defect images.

[0040] Step S2, use the YOLO neural network to perform defect image segmentation on the infrared non-destructive test image obtained in Step S1, and obtain the value of each pixel point of the mask in the image segmentation result.

[0041] Step S21: First, create a dataset of infrared non-destructive defects. The defects in the dataset are framed and marked with polygons, and the polygons should be slightly larger than the defect boundaries.

[0042] Step S22: Then, input the dataset into the YOLO image segmentation neural network for training and obtain the trained neural network.

[0043] Step S23: Finally, use the trained neural network to perform image segmentation on the infrared non-destructive test image of the standard test block obtained in Step S1. The image segmentation result will contain the mask of the defect. The essence of the mask is a matrix, and the matrix contains the values of each mask pixel point.

[0044] Step S3: Control the generated size of the mask by setting the mask threshold, and calibrate the mask threshold using the defects with known areas to obtain the calibrated mask threshold.

[0045] Step S31: Remove the mask pixel points smaller than the threshold by setting the mask threshold to control the generated size of the final mask.

[0046] Based on the matrix of the defect mask obtained in Step S2, when the value in the matrix is smaller than the set defect threshold, this point is removed, and the generated defect mask becomes smaller. Among them, when making the defect dataset in Step S2, the purpose of making the polygon framing the defect slightly larger than the defect boundary is to make the defect mask recognized by the neural network model slightly larger, which can avoid the recognized mask being smaller than the actual size of the defect, and at the same time, it also gives the defect mask room to shrink, ensuring the feasibility of correcting the defect area.

[0047] Step S32: Calibrate the mask threshold using the defects with known areas to obtain the calibrated mask threshold.

[0048] When a certain mask threshold is set, count the number of mask pixel points and multiply it by the actual area of the pixel point to obtain the defect area under this threshold. After traversing the mask thresholds within a certain range, obtain the defect area under each threshold, and the threshold with the smallest error from the actual defect area is the calibrated mask threshold.

[0049] Step S4: Under the same infrared non-destructive test conditions, obtain the infrared non-destructive test image of the workpiece to be detected, input it into the neural network and apply the calibrated mask threshold to obtain the accurate defect boundary.

[0050] Step S41: Under the same infrared non-destructive test conditions as in Step S1, obtain the infrared non-destructive test image of the workpiece to be detected.

[0051] Among them, the detection parameters include detection distance and angle, ambient temperature, acquisition frequency, exposure time, focal length, field of view, filtering parameters, imaging parameters, etc.

[0052] Step S42: Input the acquired infrared non-destructive testing image into a neural network for prediction, and apply the calibrated mask threshold. If there are defects, an accurate defect segmentation mask can be obtained.

[0053] Step S5: Calculate the number of mask pixel points and multiply it by the actual area of the pixel points to obtain the accurate defect area, thereby realizing the quantitative analysis of the internal defects of the carbon fiber honeycomb sandwich composite material.

[0054] Based on the defect mask obtained in Step S4, calculate the number of pixel points in the defect mask and multiply it by the area of the actual pixel points to obtain the actual area of the defect, thus realizing the quantitative analysis of the defect.

[0055] Embodiment

[0056] This embodiment is a method for quantitative analysis of internal defects of carbon fiber honeycomb sandwich composite materials based on artificial intelligence. The test specimen to be detected is a carbon fiber honeycomb sandwich composite material. The carbon fiber skin uses M40J, the panel and the honeycomb core are bonded with J-78B adhesive film, the skin thickness is 0.5 mm, and the honeycomb core splicing uses J-78D foaming adhesive; the height of the honeycomb core is 15 mm; a perforated honeycomb core is used, and the honeycomb core specification is LF2Y, 0.03 mm × 5 mm.

[0057] The method for fabricating the artificial prefabricated defect standard specimen is as follows: Referring to the defect requirements of GJB1719-93 "General Specification for Aluminum Honeycomb Sandwich Structures", the main internal defect types that may occur during the manufacturing process of carbon fiber sandwich composite materials are pores. The film missing method can be used for simulation. Remove the adhesive film corresponding to the area of the prefabricated defect, and apply a layer of mold release agent in the film missing area. The defect sizes are and , which are respectively used for the calibration and testing of the quantitative analysis of internal defects of carbon fiber honeycomb sandwich composite materials.

[0058] As Figure 1 shown, the method for quantitative analysis of internal defects of the carbon fiber honeycomb sandwich based on artificial intelligence in the embodiment of the present invention includes the following steps:

[0059] Step S1: Use an infrared non-destructive testing system to detect a standard test block with a defect size of . The infrared non-destructive testing system generally includes: an infrared thermal excitation module, an infrared thermal image acquisition module, and a control and post-processing module.

[0060] In this embodiment, the infrared acquisition module needs to face the standard test block directly to ensure that the workpiece and defects in the infrared image do not distort and deform, and to avoid changes in the shape and proportion of the workpiece and defect images.

[0061] Step S2: Before image segmentation, the YOLO neural network needs to be trained using existing infrared data. There are a large number of infrared non-destructive testing images in a certain carbon fiber honeycomb sandwich composite material manufacturing factory. First, use these images to create a dataset. Use the X-AnyLabeling software to box and label the defects in the dataset. Select the polygon method for boxing the defects, and the polygon should be slightly larger than the defect boundary. Then divide the created dataset into a training set, a validation set, and a test set in the ratio of 7:1.5:1.5.

[0062] Next, build the YOLO neural network. Use Python version 3.10 and the neural network framework ultralytics 8.3.25. After building the YOLO neural network environment using Anaconda and VS Code, select the yolov8m-segment neural network version for training and obtain the trained best.pt neural network parameters.

[0063] Finally, input the infrared non-destructive testing images obtained in Step S1 into the trained neural network. After image segmentation of the defects, output the defect mask matrix and the number of pixel points of the defect mask.

[0064] Step S3: Define the mask matrix in Step S2 as and convert the values in the mask matrix to values between 0 and 1. The converted matrix is represented by . The conversion formula is as follows:

[0065] ;

[0066] where represents the index of the th row and the th column in the matrix.

[0067] When the mask threshold is set to MaskThreshold, the finally generated mask matrix is as follows:

[0068] ;

[0069] In SegmentMask, 1 indicates that a mask is generated, and 0 indicates that no mask is generated; the number of pixel points for mask generation can be controlled by the mask threshold MaskThreshold.

[0070] Input the infrared image with a defect size of in Step S1 into the neural network, count the number of pixel points under different mask thresholds, and multiply by the true pixel point area of 0.8721 mm 2 to obtain the defect area under different mask thresholds, asFigure 2 As shown, it presents the law that the larger the mask threshold is, the smaller the defect area is.

[0071] Figure 2 The mask threshold range in it is from 0 to 0.99, with an interval of 0.01. The defect size is The true area is 1256.6 mm 2 , and the point closest to the true value is denoted as B. By calculating the minimum adjacent points, the coordinates of point B in Figure 2 are (0.66, 1255.68), then the calibrated mask threshold MaskThreshold is 0.66.

[0072] Step S4: Under the same detection conditions as in step S1, obtain the infrared non-destructive testing image with a defect size of , input it into the trained neural network, and use the calibrated mask threshold of 0.66 to control the size of the mask generation.

[0073] Step S5: Based on the number of pixel points of the mask in step S4, multiply it by the actual pixel point area to obtain The area of the defect is 312.30 mm 2 , and calculate according to the circular area formula The area of the defect is 314.16 mm 2 , and the error is -0.59%.

[0074] Figure 3 is The original infrared non-destructive testing image of the defect and the masks under different mask thresholds. It can be seen that the larger the mask threshold is, the smaller the mask area is, which proves the necessity of mask threshold calibration.

[0075] When the mask threshold is set to 0.1, the obtained defect area is 361.94 mm 2 , and the error is 15.21%; when the mask threshold is set to 0.9, the obtained defect area is 259.22 mm 2 , and the error is -17.49%.

[0076] When the mask threshold setting deviates from the calibrated value, the defect area error is relatively large, while the error after calibration is only -0.59%, which further proves that the quantitative analysis method of calibrating first and then measuring is necessary and effective.

[0077] Among them, a large number of infrared non-destructive testing images of the carbon fiber honeycomb sandwich composite to be detected can be obtained in step S4. After performing the quantitative analysis in step S5, the tasks of rapid, large-batch detection and defect quantitative analysis can be realized, providing technical support for the quality control of the composite material.

[0078] Therefore, the present invention adopts the above-mentioned method for quantitative analysis of internal defects of carbon fiber honeycomb sandwich based on artificial intelligence. By using infrared non-destructive testing technology combined with artificial intelligence algorithms, defect image segmentation of infrared images is performed through the YOLO neural network. After using the calibration method of the mask threshold, more accurate quantitative analysis of defects can be achieved. By using the artificial intelligence method, only the calibrated threshold needs to be set, and under the same detection conditions, rapid and large-scale detection and defect quantitative analysis tasks can be realized, providing technical support for the quality control of composite materials.

[0079] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that they can still modify or equivalently replace the technical solutions of the present invention, and these modifications or equivalent replacements do not make the modified technical solutions deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A quantitative analysis method for internal defects of carbon fiber honeycomb sandwich based on artificial intelligence, characterized in that: The following steps are involved: Step S1, obtaining an infrared nondestructive testing image of a standard test block of a carbon fiber sandwich composite material with a known defect area; Step S2, using the YOLO neural network to perform defect image segmentation on the infrared nondestructive testing image obtained in step S1, and obtaining the value of each pixel point of the mask in the image segmentation result; Step S3, control the size of the mask by setting the mask threshold, calibrate the mask threshold using a defect of known area, and obtain the calibrated mask threshold. The specific process is as follows: Step S31, based on the matrix of the defect mask obtained in step S2, a mask threshold is set, and when the value in the matrix is ​​less than the set defect threshold, mask pixels less than the threshold are removed to control the size of the final mask generated; Step S32, calibrating the mask threshold using a defect of known area to obtain a calibrated mask threshold; After setting a certain mask threshold, count the number of mask pixels and multiply it by the actual area of ​​the pixels to get the defect area under the threshold; after traversing the mask thresholds within the range, get the defect area under each threshold, and the threshold with the smallest error with the actual defect area is the calibrated mask threshold; Step S4, under the same infrared nondestructive testing conditions, obtaining an infrared nondestructive testing image of the workpiece to be tested, inputting it into the neural network and applying the calibrated mask threshold to obtain an accurate defect boundary; Step S5: Calculate the number of mask pixels and multiply it by the actual area of ​​the pixels to obtain an accurate defect area, thereby achieving quantitative analysis of internal defects of the carbon fiber honeycomb sandwich composite material.

2. The method for quantitative analysis of internal defects of carbon fiber honeycomb sandwich based on artificial intelligence according to claim 1, characterized in that: In step S1, an infrared nondestructive testing device is used to test a standard test block with a known defect area, and an infrared nondestructive testing image is obtained; When acquiring infrared nondestructive testing images, the infrared nondestructive testing equipment should be used to test the standard test block to ensure that the acquired workpiece and defect images are not distorted or deformed, and to avoid changes in the shape and proportion of the workpiece and defect images.

3. The method for quantitative analysis of internal defects of carbon fiber honeycomb sandwich based on artificial intelligence according to claim 2, characterized in that: The infrared nondestructive testing equipment includes an infrared thermal excitation module, an infrared thermal image acquisition module, and a control and post-processing module.

4. The method for quantitative analysis of internal defects of carbon fiber honeycomb sandwich based on artificial intelligence according to claim 1, characterized in that: In step S2, the YOLO neural network is used to perform defect image segmentation on the infrared nondestructive testing image obtained in step S1, and the value of each pixel of the mask in the image segmentation result is obtained. The specific process is as follows: Step S21, first, a data set of infrared non-destructive defects is prepared, and the defects in the data set are framed and marked with polygons, and the polygons are larger than the defect boundaries; Step S22, then, input the data set into the YOLO image segmentation neural network, perform training and obtain the trained neural network; Step S23, finally, use the trained neural network to perform image segmentation on the infrared nondestructive testing image of the standard test block obtained in step S1, and the image segmentation result will include a mask of the defect; The essence of the mask is a matrix, which contains the value of each mask pixel.

5. The method for quantitative analysis of internal defects of carbon fiber honeycomb sandwich based on artificial intelligence according to claim 1, characterized in that: In step S4, under the same infrared nondestructive testing conditions, the infrared nondestructive testing image of the workpiece to be tested is obtained, input into the neural network and the calibrated mask threshold is applied to obtain the accurate defect boundary. The specific process is as follows: Step S41, under the same infrared nondestructive testing conditions as step S1, obtaining an infrared nondestructive testing image of the workpiece to be tested; Among them, the detection parameters include detection distance and angle, ambient temperature, acquisition frequency, exposure time, focal length, field of view, filtering parameters, and imaging parameters; Step S42: input the acquired infrared nondestructive testing image into the neural network for prediction, and apply the calibrated mask threshold to obtain an accurate defect segmentation mask if defects exist.

6. The method for quantitative analysis of internal defects of carbon fiber honeycomb sandwich based on artificial intelligence according to claim 1, characterized in that: In step S5, the defect mask is obtained according to step S4, the number of mask pixels is calculated, and the number is multiplied by the actual area of ​​the pixels to obtain the accurate defect area, thereby achieving quantitative analysis of internal defects of the carbon fiber honeycomb sandwich composite material.

Images