Composite material bending plate defect identification method and system and readable storage medium
By using machine learning and ultrasonic guided wave technology, a convolutional neural network model was constructed, which solved the problems of numerous blind spots and difficulty in distinguishing defects in the bending area of composite material plates, and achieved automatic, fast and accurate defect identification.
Patent Information
- Application Number
- CN202210811552.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-07-11
- Publication Date
- 2026-02-27
- Estimated Expiration
- 2042-07-11
AI Technical Summary
Defect detection in the bending area of composite material bending plates has many blind spots and is difficult to distinguish, making it difficult to achieve rapid and accurate identification.
By employing machine learning and ultrasonic guided wave technology, composite material bent plate samples are fabricated, ultrasonic guided wave signals are collected and converted into time-frequency images, and a convolutional neural network model is constructed for training to achieve defect identification.
It enables automatic, rapid, and accurate identification of defects in composite material bending plates, improving detection efficiency and accuracy, and allowing for qualitative and quantitative identification of defects.
Smart Images
Figure CN115166047B_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application belongs to the technical field of composite material structure damage detection, and particularly relates to a composite material bending plate defect identification method, system and readable storage medium. BACKGROUND
[0002] Composite materials are becoming the preferred material in the aerospace, naval vessels and automotive industries due to their high specific strength, high specific stiffness, strong design flexibility, good damping performance and other advantages. In particular, in aerospace structural components, complex-shaped composite material components such as composite material bending plates are gradually replacing traditional metal material components.
[0003] However, the bending area of the composite material bending plate is susceptible to stress concentration, fatigue load and other factors, which may cause defects such as delamination, holes and matrix cracking in the bending area. These defects can seriously affect the strength, stiffness and service life of the composite material bending plate, and even cause catastrophic failure of the composite material structure. Therefore, in order to ensure the safe and reliable operation of the composite material bending plate in actual engineering structures, it is necessary to identify and quantify these defects in the composite material bending plate in a timely manner. However, when detecting defects in the bending area of the composite material bending plate, there are many blind spots and it is difficult to distinguish, which brings great difficulty to the defect detection of the bending area of the composite material bending plate. SUMMARY
[0004] Therefore, the present application provides a composite material bending plate defect identification method, which is based on machine learning and ultrasonic guided wave technology and can quickly and accurately identify defects in the bending area of the composite material bending plate.
[0005] To achieve the above-mentioned purpose, the specific scheme adopted by the present application is as follows:
[0006] The composite material bending plate defect identification method comprises the following steps:
[0007] Step 1: Make a composite material bending plate sample, wherein the bending area of the composite material bending plate sample has a healthy area and two or more different defect areas;
[0008] Step 2: Collect ultrasonic guided wave signals of the healthy area and the defect area of the bending area of the composite material bending plate sample, and give labels to the ultrasonic guided wave signals of the healthy area and the defect area, respectively;
[0009] Step 3: Convert the ultrasonic guided wave signals in step 2 into corresponding time-frequency images;
[0010] Step 4: Based on the time-frequency images in step 3, a convolutional neural network model is constructed;
[0011] Step five: taking the time-frequency image of step three as the input of the convolutional neural network model in step four, and taking the corresponding label in step two as the output, training the convolutional neural network model to obtain a composite material bending plate defect identification model;
[0012] Step six: inputting the time-frequency image corresponding to the ultrasonic guided wave signal of the composite material bending plate to be tested into the composite material bending plate defect identification model in step five, and identifying the defects of the composite material bending plate according to the label output by the composite material bending plate defect identification model.
[0013] Further, in step three, the Hamming window function in the short-time Fourier transform method is used to intercept the ultrasonic guided wave signal to obtain the time-frequency image of the ultrasonic guided wave signal.
[0014] Further, the convolutional neural network model in step four includes three convolutional layers, three pooling layers and three fully connected layers.
[0015] Further, the time-frequency image data of the ultrasonic guided wave signal in step three is divided into a training set and a test set.
[0016] The training set is used for training the convolutional neural network model in step five.
[0017] The test set is used to verify the composite material bending plate defect identification model in step five.
[0018] Further, the batch normalization method is used to train the convolutional neural network model in step five.
[0019] Further, when collecting the ultrasonic guided wave signals of the healthy area and the defect area of the composite material bending plate sample in step two, the excitation transducer and the receiving transducer are arranged on two planes of the composite material bending plate sample, and the defect area and the healthy area are located between the excitation transducer and the receiving transducer.
[0020] Further, when collecting the ultrasonic guided wave signals of the healthy area and the defect area of the composite material bending plate sample in step two, the coupling condition is changed for multiple times of collection.
[0021] Further, in step one, two or more defects of different types and / or two or more defects of the same type but different sizes are arranged in the composite material bending plate sample.
[0022] Moreover, the present application also provides a composite material bending plate defect identification system, which can automatically identify the defect condition corresponding to the ultrasonic guided wave signal after collecting the ultrasonic guided wave signal of the bending area of the composite material, and the specific technical solutions adopted are as follows:
[0023] A composite bending plate defect identification system comprises:
[0024] A data acquisition module is configured to acquire ultrasonic guided wave signals of the composite bending plate at a bending portion and send the signals to a time-frequency image generation module;
[0025] The time-frequency image generation module is configured to generate corresponding time-frequency images of the ultrasonic guided wave signals and send the images to a defect identification module;
[0026] The defect identification module is internally preset with a composite bending plate defect identification model, receives the time-frequency images and inputs the images into the composite bending plate defect identification model, and identifies defects at the bending portion of the composite bending plate according to an output of the composite bending plate defect identification model.
[0027] In addition, the application further provides a readable storage medium, and a computer program is stored on the storage medium, and the computer program is executed to perform the steps of the composite bending plate defect identification method.
[0028] Advantages:
[0029] (1) The composite bending plate defect identification method of the application is based on machine learning and ultrasonic guided wave technology, converts one-dimensional ultrasonic guided wave signals into two-dimensional time-frequency images, extracts and classifies features of the time-frequency images by using a neural network, obtains a composite bending plate defect identification model, and can automatically, quickly and accurately identify defects of the composite bending plate by using the composite bending plate defect identification model.
[0030] (2) The composite bending plate defect identification method of the application uses a Hamming window function in short-time Fourier transform to process the acquired ultrasonic guided wave signals of the composite bending plate, and obtains two-dimensional time-frequency images with high resolution.
[0031] (3) The composite bending plate defect identification method of the application constructs a convolutional neural network model comprising an input layer, three convolutional layers, three pooling layers, three fully connected layers and an output layer, a total of eleven layers, and the increase in the number of network layers helps to improve the network performance and enhance the recognition speed and accuracy of the composite bending plate defect model.
[0032] (4) In the composite bending plate defect identification method of the application, the excitation transducer and the receiving transducer are arranged on two different planes of the composite bending plate sample when the ultrasonic guided wave signals of the healthy region and the defect region of the bending portion of the composite bending plate sample are collected, and the defect region and the healthy region are located between the excitation transducer and the receiving transducer, so that the entire region (not limited to the bending region) between the two transducers can be detected each time, and the detection efficiency is high.
[0033] (5) The composite material bending plate defect identification method of the present application sets different types and sizes of defects at the bending part of the composite material bending plate, so that the composite material bending plate defect identification model can both qualitatively and quantitatively identify the defects at the bending part of the composite material.
[0034] (6) The composite material bending plate defect identification system of the present application comprises a data acquisition module, a time-frequency image generation module and a defect identification module, and through the system, the defects at the bending part of the composite material bending plate can be automatically, quickly, efficiently and accurately identified. BRIEF DESCRIPTION OF DRAWINGS
[0035] Figure 1 The flowchart of the composite material bending plate defect intelligent identification method based on machine learning and ultrasonic guided wave provided by the first embodiment of the present application;
[0036] Figure 2 The structure schematic diagram of the composite material bending plate sample provided by the second embodiment of the present application, wherein (a) is a layered defect sample, and (b) is a flat-bottom hole defect sample;
[0037] Figure 3 The structure schematic diagram of the convolutional neural network model provided by the second embodiment of the present application;
[0038] Figure 4 The schematic diagram of the installation position of the transducer when the ultrasonic guided wave signal of the bending area of the composite material bending plate sample is collected in the second embodiment of the present application. DETAILED DESCRIPTION
[0039] The present application will be described in detail below with reference to the accompanying drawings and embodiments.
[0040] Embodiment one:
[0041] Ultrasonic guided wave has the advantages of long propagation distance and high detection efficiency, and is widely used in the detection of defects of plate-shaped structures. When ultrasonic guided wave encounters a defect on the propagation path, scattering, reflection and other phenomena occur, which in turn causes abnormal changes in the shape and energy of the guided wave signal. The type, position and size of the defect can be judged by the characteristics of the abnormal signal. However, if ultrasonic guided wave is used for composite material bending plate defect detection, the frequency dispersion and complex modal conversion phenomenon of the guided wave signal, the anisotropy of the composite material and the reflection and scattering of the guided wave signal at the bending area of the bending plate make it difficult to analyze the guided wave signal during bending plate defect detection, and the defect cannot be accurately identified.
[0042] The embodiment provides a composite material bent plate defect intelligent identification method based on machine learning and ultrasonic guided waves. The method first collects guided wave signals, then converts one-dimensional guided wave signals into two-dimensional time-frequency images by using short-time Fourier transform, designs a convolutional neural network model, trains the convolutional neural network by using the time-frequency images and labels, performs autonomous feature extraction and classification of defects based on the convolutional neural network, and finally predicts the defects by using the trained convolutional neural network, so that accurate identification of the defects is realized.
[0043] A flowchart of the composite material bent plate defect identification method is shown in Figure 1 The method comprises the following steps:
[0044] Step 1: A composite material bent plate sample is prepared, and the bent part of the composite material bent plate sample has a healthy area and two or more different defect areas.
[0045] Step 2: Ultrasonic guided wave signals of the healthy area and the defect area of the bent part of the composite material bent plate sample are collected, and the ultrasonic guided wave signals of the healthy area and the defect area are respectively given labels.
[0046] Step 3: The ultrasonic guided wave signals in step 2 are converted into corresponding time-frequency images.
[0047] Step 4: Based on the time-frequency images in step 3, a convolutional neural network model is constructed.
[0048] Step 5: The time-frequency images in step 3 are used as the input of the convolutional neural network model in step 4, and the corresponding labels in step 2 are used as the output, the convolutional neural network model is trained, and a composite material bent plate defect identification model is obtained.
[0049] Step 6: The time-frequency images corresponding to the ultrasonic guided wave signals of the composite material bent plate to be tested are input into the composite material bent plate defect identification model, and the labels output by the composite material bent plate defect identification model are used to identify the defects of the composite material bent plate.
[0050] Embodiment 2
[0051] On the basis of the above-mentioned embodiment 1, the embodiment provides a method for identifying defects of a specific bent part of a composite material bent plate.
[0052] Step A
[0053] A composite material bent plate sample is prepared, and two or more different types and / or two or more different sizes of the same type of defects are arranged in the composite material bent plate sample. For example, Figure 2As shown in (a) and (b), two composite bending plate specimens were fabricated. The two composite bending plate specimens have 16 layers, each with a thickness of 0.1875 mm. The material is M55J carbon fiber prepreg, and the layup sequence is [(0° / 45° / 90° / -45°)2]. S Furthermore, the dimensions of both flat plates in these two composite material curved plate samples are 300mm in length × 300mm in width × 3mm in thickness. Figure 2 In (a) and (b), delamination defects and flat-bottom hole defects were prefabricated in the bending areas of the composite material bending plate specimens. The delamination defects were simulated by pre-embedding polytetrafluoroethylene with diameters of Φ6mm, Φ10mm, and Φ15mm and a thickness of 0.1mm at half the thickness of the bending area of the composite material bending plate specimen. The flat-bottom hole defects had diameters of Φ2mm, Φ7mm, and Φ10mm and a depth of 1.5mm.
[0054] Step B:
[0055] Acquire ultrasonic guided wave signals from healthy and defective regions at the bending point of composite material bent plate specimens and assign corresponding labels.
[0056] Taking the acquisition of ultrasonic guided wave signals from the healthy region as an example, guided waves are excited on one side of the healthy region at the bending point of the composite material bent plate specimen, and guided wave signals are acquired on the other side to obtain the corresponding ultrasonic guided wave signals. The same method is used to acquire ultrasonic guided wave signals from the defective region.
[0057] like Figure 3 As shown, when acquiring ultrasonic guided wave signals at the bending point of the composite material bent plate specimen, the excitation transducer excites ultrasonic guided waves on one side of a healthy area and three defective areas (including delamination defects and flat-bottom hole defects) at the bending point of the composite material bent plate specimen, and the receiving transducer acquires ultrasonic guided wave signals on the other side. More specifically, the excitation signal is a 5-cycle sine wave with a center frequency of 5MHz. More specifically, as shown... Figure 3 As shown, the ultrasonic guided wave propagation distance between the two transducers (excitation transducer and receiving transducer) is 100 mm, and the defect area and the healthy area are located 50 mm in the middle of the ultrasonic guided wave propagation path. Furthermore, each of the three defect areas and the healthy area acquires 200 ultrasonic guided wave signals by changing the coupling conditions, thus obtaining a total of 800 ultrasonic guided wave signals on each sample. The ultrasonic guided wave signal labels for the healthy area and the three defect areas on each composite material bent plate sample are set to 0, 1, 2, and 3, respectively. Taking delamination defects as an example, 0: healthy, 1: Φ6 delamination defect, 2: Φ10 delamination defect, 3: Φ15 delamination defect.
[0058] Step C:
[0059] The ultrasonic guided wave signal provided with the label is processed by using a short-time Fourier transform method to obtain a time-frequency image corresponding to the ultrasonic guided wave signal.
[0060] In the embodiment, for an energy-limited time-domain signal x(t), a short-time window function h(t) is loaded on the basis of Fourier transform. With the movement of the window function, Fourier transform is performed on the time-domain signal in each window function to obtain a short-time Fourier transform result. In order to achieve the best resolution, the Hamming window function is used to intercept the ultrasonic guided wave signal, and then fast Fourier transform is performed. The window function is slid from left to right, and fast Fourier transform is sequentially performed. Finally, the fast Fourier transform of each segment of the ultrasonic guided wave signal is converted into a two-dimensional time-frequency image. The original size of each time-frequency image is 875*656*3 (width*height*channel number). In order to reduce the calculation cost of the machine learning algorithm, the image size is adjusted and center cropping is performed. The size after cropping is 46*46*3.
[0061] The ultrasonic guided wave signal is processed by using the short-time Fourier transform to convert the one-dimensional guided wave signal into a two-dimensional time-frequency image. The time and frequency information can be obtained simultaneously. Compared with separate time-domain analysis and frequency-domain analysis, the short-time Fourier transform method can simultaneously describe the energy changes in the time and frequency two scales, thereby representing the local features of the signal in the time domain and the frequency domain, which is helpful to mine the defect information in the ultrasonic guided wave signal.
[0062] Step D:
[0063] A convolutional neural network model is constructed based on the time-frequency image in step C. Specifically, a convolutional neural network model is constructed on a PyTorch deep learning framework according to the characteristics of the input image. The model is composed of an input layer, three convolutional layers, three pooling layers, three fully connected layers, and an output layer, as shown in Figure 4 The convolutional neural network model is constructed based on the classic LeNet-5 model. However, the LeNet-5 model has problems such as slow convergence speed, low calculation efficiency, weak generalization ability, and low classification accuracy. Therefore, the convolutional neural network model is improved in the following ways:
[0064] 1) The input layer of LeNet-5 has an image size of 32*32 black and white image. The convolutional neural network model uses a 46*46*3 color image. Compared with black and white images, color images can provide more information;
[0065] 2) The convolution kernel size of LeNet-5 is 5*5. The convolutional neural network model uses a 3*3 convolution kernel and increases the number of convolution kernels. Using smaller and more convolution kernels helps to better extract features;
[0066] 3) The number of network layers is increased from 8 to 11, and the increase in the number of network layers helps to improve network performance;
[0067] 4) Batch normalization layer is used after convolution layer, which can accelerate network convergence and improve network generalization ability;
[0068] 5) Relu activation function is used instead of Sigmoid activation function, which can effectively alleviate the gradient disappearance problem and improve the calculation speed.
[0069] Step E:
[0070] The time-frequency image in step C is taken as the input of the convolutional neural network model in step D, and the corresponding label in step B is taken as the output. The convolutional neural network model is trained to obtain a composite material bending plate defect recognition model. Specifically, the processed (such as data cleaning and data feature scaling) ultrasonic guided wave signal data is divided into a training set and a test set, the proportion of the training set is 70%, and the proportion of the test set is 30%. The time-frequency image and the corresponding label of the training set are used to train the convolutional neural network model, and the batch normalization method is used to speed up the training speed and improve the accuracy of the composite material bending plate defect recognition model. The convolutional neural network weights are randomly initialized, and the Adam optimizer is used for calculation. After the convolutional neural network converges, the composite material bending plate defect recognition model is obtained. Then the test set is used to verify the composite material bending plate defect recognition model, and the error on the test set is taken as the generalization error of the final model in the real scene.
[0071] Step F:
[0072] The time-frequency image corresponding to the ultrasonic guided wave in the carbon fiber composite material bending plate to be tested is input into the composite material bending plate defect recognition model in step E, and the label output by the composite material bending plate defect recognition model is used to identify the defects at the bending position of the composite material bending plate. It is worth noting that if the actual sample defect is different from the Φ6 delamination defect, the Φ10 delamination defect and the Φ15 delamination defect on the sample, then the composite material bending plate defect recognition model will be classified into one of the defects according to the actual defect and the Φ6 delamination defect, the Φ10 delamination defect and the Φ15 delamination defect.
[0073] Specifically, for the trained composite bending plate defect recognition model, the time-frequency image corresponding to the ultrasonic guided wave in the test set is taken as the input, and the composite bending plate defect recognition model automatically calculates and outputs the corresponding label. Different labels represent different defects. For example, 0: healthy, 1: Φ6 delamination defect, 2: Φ10 delamination defect, and 3: Φ15 delamination defect. Thus, the defect recognition result can be obtained.
[0074] The present embodiment provides two groups of experimental data for processing, including two different types of defects. After the above steps are performed on the data of each sample, the defect recognition results of the two composite bending plate samples are shown in Table 1.
[0075] Table 1 Defect recognition results of composite bending plate samples
[0076]
[0077] It can be seen that an accuracy of more than 99.2% is obtained on the two samples, realizing accurate recognition of the defects of the composite bending plate and verifying the effectiveness of the method.
[0078] The composite bending plate defect recognition method uses a convolutional neural network to extract features and classify time-frequency images. The convolutional neural network has excellent image recognition and classification capabilities. The convolutional neural network automatically extracts image features, overcoming the inaccuracy and tediousness of manual feature extraction. It can effectively extract defect-related features in the time-frequency image of the guided wave signal and improve the accuracy of defect recognition. Moreover, by using short-time Fourier transform to convert one-dimensional guided wave signals into two-dimensional time-frequency images, the convolutional neural network automatically extracts defect-related features in the time-frequency image and automatically classifies defects, so that the method no longer relies on manual judgment.
[0079] Embodiment Three
[0080] Based on the above-mentioned embodiments one and two, the present embodiment provides a composite bending plate defect recognition system. The system comprises:
[0081] The data acquisition module is used to acquire the ultrasonic guided wave signal of the composite bending plate at the bending part and send it to the time-frequency image generation module.
[0082] The time-frequency image generation module is used to generate the corresponding time-frequency image of the ultrasonic guided wave signal and send it to the defect recognition module.
[0083] The defect recognition module has a composite bending plate defect recognition model preset inside. The received time-frequency image is input into the composite bending plate defect recognition model, and the defects at the bending part of the composite bending plate are recognized according to the output of the composite bending plate defect recognition model.
[0084] Embodiment Four
[0085] On the basis of the above-mentioned embodiment one, the embodiment provides a readable storage medium, which stores a computer program, and the computer program can realize the steps of the method in the embodiment one when executed by a processor.
[0086] To sum up, the above is only a preferred embodiment of the present application, not for limiting the protection scope of the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.
Claims
1. A method for identifying defects in composite material bent plates, characterized in that, Includes the following steps: Step 1: Prepare composite material bent plate specimens, wherein the bent plate specimens have healthy areas and two or more different defect areas at the bending point; Step 2: Collect ultrasonic guided wave signals from the healthy area and the defective area at the bending point of the composite material bending plate sample, and assign labels to the ultrasonic guided wave signals from the healthy area and the defective area respectively; Step 3: Convert the ultrasonic guided wave signal described in Step 2 into a corresponding time-frequency image; Step 4: Based on the time-frequency images from Step 3, construct a convolutional neural network model; Step 5: Use the time-frequency image described in Step 3 as the input to the convolutional neural network model in Step 4, and the corresponding label from Step 2 as the output to train the convolutional neural network model and obtain a composite material bending plate defect recognition model. Step 6: Input the time-frequency image corresponding to the ultrasonic guided wave signal of the composite material bending plate to be tested into the composite material bending plate defect identification model in Step 5, and identify the defects at the bending point of the composite material bending plate according to the labels output by the composite material bending plate defect identification model. Step 3 uses the Hamming window function in the short-time Fourier transform method to truncate the ultrasonic guided wave signal and obtain the time-frequency image of the ultrasonic guided wave signal; The convolutional neural network model described in step four is built based on the LeNet-5 model and uses the ReLU activation function instead of the Sigmoid activation function. It includes three convolutional layers, three pooling layers, and three fully connected layers. In step two, when collecting ultrasonic guided wave signals from the healthy region and the defective region at the bending point of the composite material bent plate sample, the excitation transducer and the receiving transducer are respectively set on two planes of the composite material bent plate sample, and the defective region and the healthy region are located between the excitation transducer and the receiving transducer. In step two, when collecting ultrasonic guided wave signals from the healthy and defective regions at the bending point of the composite material bent plate sample, multiple acquisitions are performed by changing the coupling conditions. Step 5 involves training the convolutional neural network model using a batch normalization method, randomly initializing the convolutional neural network weights, and using the Adam optimizer for computation.
2. The method for identifying defects in composite material bent plates as described in claim 1, characterized in that, The time-frequency image data of the ultrasonic guided wave signal mentioned in step three is divided into a training set and a test set; The training set was used in step five to train the convolutional neural network model; The test set is used to validate the composite material bending plate defect identification model in step five.
3. The method for identifying defects in composite material bent plates as described in any one of claims 1 to 2, characterized in that: In step one, two or more defects of different types and / or two or more defects of the same type but different sizes are set in the composite material bending plate specimen.
4. A composite material bending plate defect identification system, used to implement the composite material bending plate defect identification method according to any one of claims 1 to 3, characterized in that, include: Data acquisition module: used to acquire ultrasonic guided wave signals at the bending point of composite material bent plates and send them to the time-frequency image generation module; Time-frequency image generation module: used to generate a corresponding time-frequency image from the ultrasonic guided wave signal and send it to the defect identification module; Defect identification module: It has a pre-set composite material bending plate defect identification model; it inputs the received time-frequency image into the composite material bending plate defect identification model, and identifies the defects at the bending point of the composite material bending plate according to the output of the composite material bending plate defect identification model.
5. A readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 3.