Ultrasonic guided wave damage location method for resin-based carbon fiber composite material plate
Patent Information
- Application Number
- CN202311375432.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-10-23
- Publication Date
- 2026-09-15
- Estimated Expiration
- 2043-10-23
AI Technical Summary
然而，在制造到加工成型的过程中，或在服役工作状态及外部环境的影响下，结构的表面及内部不可避免地会产生微小的缺陷，而且这些缺陷还会逐步扩展和合并，形成宏观的损伤，甚至造成重大的灾难性事故
[0037] 1. This invention proposes an ultrasonic guided wave damage localization method for resin-based carbon fiber composite plates. A sparse sensor guided wave matrix is constructed, and the simulated damage guided wave array signal is obtained by Fourier transform. Simulated damage detection experiments are conducted, and a Long Short-Term Memory (LSTM) network model is constructed using LSTM to successfully locate the damage position.
Smart Images
Figure CN117420216B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of nondestructive testing technology, and in particular, it is a method for locating ultrasonic guided wave damage in resin-based carbon fiber composite plates. Background Technology
[0002] Resin-based carbon fiber composites possess high tensile strength, excellent corrosion resistance, and high impact resistance, making them widely used in the aerospace field. With the rapid development of the aerospace industry, the demand for resin-based carbon fiber composite sheets is increasing daily, particularly for components such as aircraft wings and fuselages. However, during manufacturing and processing, or under the influence of service conditions and the external environment, microscopic defects inevitably develop on the surface and internal structure. These defects can gradually expand and merge, forming macroscopic damage and even causing major catastrophic accidents.
[0003] Lamb waves are elastic guided waves formed by the superposition of ultrasonic waves reflected from two free boundaries in a thin plate structure. In the fields of non-destructive testing and structural health monitoring, compared to traditional transverse and longitudinal waves, Lamb waves are widely used due to their advantages such as boundary reflection, low attenuation, long detection distance, high detection efficiency, and sensitivity to damage. The time-of-flight difference (TOF), the time difference between the direct wave and the damage-reflected wave in the time domain signal, is the most intuitive damage characteristic parameter. It clearly reflects the positional relationship between the excitation sensor, the receiving sensor, and the damage, providing a solid theoretical basis for damage localization. The arrangement of array sensors allows the signal to contain spatial information; for example, the elliptic localization method uses multiple sensors and the TOF to locate the damage. Deep learning can automatically learn useful features from large amounts of data, including the spatiotemporal characteristics of Lamb wave signals, achieving damage localization and showing significant advantages over other methods. Therefore, it is urgent and necessary to find an ultrasonic guided wave damage localization method for resin-based carbon fiber composite plates by arranging array sensors to acquire spatial information of the detection area, so as to achieve rapid and accurate damage localization of resin-based carbon fiber composite plates. Summary of the Invention
[0004] This invention addresses the shortcomings of existing technologies by proposing an ultrasonic guided wave damage localization method for resin-based carbon fiber composite plates. The method includes: constructing a sparse sensor guided wave matrix to acquire guided wave array signals; establishing a damage detection simulation model and acquiring corresponding damage detection training and testing sets under simulated damage; constructing and training a Long Short-Term Memory (LSTM) network model; introducing sparse weights to obtain a second loss function; training the LSTM network model based on the damage detection training and testing sets to obtain the optimal LSTM network model; and obtaining damage localization imaging results based on measured guided wave array signals from actual damage. This invention improves the accuracy of model damage localization and reduces localization noise by constructing a sparse sensor guided wave matrix, acquiring simulated damage guided wave array signals using Fourier transform, employing an LSTM network, and improving the loss function by introducing sparse weights.
[0005] This invention provides a method for locating ultrasonic guided wave damage in a resin-based carbon fiber composite plate, comprising the following steps:
[0006] S1. Construct a sparse sensor waveguide matrix and obtain the waveguide array signal;
[0007] S2. Establish a damage detection simulation model and obtain the corresponding damage detection training set and damage detection test set under simulated damage.
[0008] S3. Construct a Long Short-Term Memory (LSTM) network model: The LSTM network model includes LSTM layers and fully connected layers; the output of the LSTM network model is a localization matrix; the first loss function of the LSTM network model is set as follows:
[0009] loss = (loc - label) 2 (2)
[0010] Where label represents the label matrix; loc represents the localization matrix output by the Long Short-Term Memory (LSTM) network model in each training cycle; the localization matrix has the same size as the label matrix;
[0011] S4. Introduce sparse weights to obtain the second loss function;
[0012] S41. Obtain the localization matrix loc output by the Long Short-Term Memory (LSTM) network model in each training cycle;
[0013] S42. Introduce sparse weights and define a second loss function. x Represented as:
[0014] loss x=loss+λ|loc| (3)
[0015] Where λ represents sparse weights; |loc| represents the modulus of the localization matrix loc output by the Long Short-Term Memory (LSTM) model in each training cycle;
[0016] S5. Based on the damage detection training set and the damage detection test set, train the Long Short-Term Memory (LSTM) network model and obtain the optimal LSTM network model.
[0017] S6. Obtain damage localization imaging results based on measured guided wave array signals of actual damage: Acquire measured guided wave array signals of actual damage, input them into the optimal Long Short-Term Memory (LSTM) network model, and use the second loss function... x The localization matrix is then output, which is the damage localization imaging result.
[0018] Furthermore, step S1 specifically includes the following steps:
[0019] S11. Construct a sparse sensor waveguide matrix and determine the detection area: Based on W sensors, construct a sparse sensor waveguide matrix and set the area covered by the sparse sensor waveguide matrix as the detection area; the W sensors are numbered 1 to W respectively;
[0020] S12. Acquire several sets of transmitting and receiving sensor groups and waveguide array signals: sequentially excite the sensors numbered 1 to W as excitation sensors. Each time the sensors are excited, W-1 sensors other than the excited sensor are used as receiving sensors, and the waveguide signals of all the receiving sensors are collected at the same time. A total of W(W-1) sets of transmitting and receiving sensor groups and the corresponding waveguide signals are obtained to form the waveguide array signal.
[0021] S13. Exclude waveguide signals with the same waveform from the waveguide array signals: In all the transmitting and receiving sensor groups, only one of the excitation sensor and the receiving sensor corresponding to the sensor is retained if they are interchanged. At the same time, only one of the corresponding waveguide array signals in the array signals is also retained.
[0022] Preferably, step S2 specifically includes the following steps:
[0023] S21. Establish a damage detection simulation model: Divide the detection area into an N×N grid and construct an N×N label matrix corresponding to the simulated damage.
[0024] S22. Calculate the distance of simulated damage propagation: Calculate the sum of the distances from the simulated damage to the excitation sensor and the receiving sensor, which is the distance the Lamb wave propagates from the excitation sensor to the damage reflection and then to the receiving sensor.
[0025] S23. The short pure tone toneburst signal g(t) is used as the excitation of the excitation sensor, and its expression is:
[0026]
[0027] Where f represents the center frequency; cyc represents the number of peaks; and t represents time.
[0028] S24. Perform a Fast Fourier Transform on g(t), and set the sampling frequency f. s M waveguide array signal samples are obtained;
[0029] S25. Obtain the waveguide array signal corresponding to the simulated damage: Based on the Lamb wave propagation group velocity and frequency-thickness product relationship curve, obtain the propagation speed of the waveguide signal in the plate at different center frequencies f, calculate the frequency domain signal of the waveguide signal when the propagation distance is D, perform inverse Fourier transform and normalization of the waveguide signal in sequence, and add white noise with an amplitude of r to obtain the waveguide array signal corresponding to the simulated damage.
[0030] S26. Construct the damage detection training set and damage detection test set: Each simulated damage location corresponds to a set of guided wave array signals and a label matrix. In each grid, a points are randomly selected to represent simulated damage, resulting in a × N × N damage detection training sets. In the entire detection area, b points are randomly selected to obtain b damage detection test sets.
[0031] Preferably, in step S12, each of the transmitting and receiving sensor groups includes one excitation sensor and one receiving sensor; the guided wave array signal includes W(W-1) groups of guided wave signals; and in step S13, the guided wave array signal after exclusion includes W(W-1) / 2 groups of guided wave signals.
[0032] Preferably, in step S21, if there is damage within the grid, the corresponding position of the label matrix is represented as 1; otherwise, the corresponding position of the label matrix is represented as 0.
[0033] Preferably, in step S3, the Long Short-Term Memory (LSTM) network has two layers; the fully connected layers have two layers; the optimizer for the LSTM network model is Adam; and the hidden layer dimension is 800.
[0034] Preferably, the hyperparameters used for training the Long Short-Term Memory (LSTM) network model in step S5 are set as follows: the optimizer is Adam and the hidden layer dimension is 800.
[0035] Preferably, the amplitude r of the white noise in step S25 is 0.3.
[0036] Compared with the prior art, the technical effects of the present invention are as follows:
[0037] 1. This invention proposes an ultrasonic guided wave damage localization method for resin-based carbon fiber composite plates. A sparse sensor guided wave matrix is constructed, and the simulated damage guided wave array signal is obtained by Fourier transform. Simulated damage detection experiments are conducted, and a Long Short-Term Memory (LSTM) network model is constructed using LSTM to successfully locate the damage position.
[0038] 2. The present invention provides an ultrasonic guided wave damage localization method for resin-based carbon fiber composite plates. By introducing sparse weights, the loss function of the Long Short-Term Memory (LSTM) network model is improved, thereby increasing the accuracy of damage localization and reducing localization noise. Attached Figure Description
[0039] Other features, objects, and advantages of this application will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings.
[0040] Figure 1 This is a flowchart of the ultrasonic guided wave damage localization method for resin-based carbon fiber composite plates of the present invention;
[0041] Figure 2 This is a schematic diagram of the sensor position and detection area in a specific embodiment of the present invention;
[0042] Figure 3 This is a schematic diagram of the relationship between the Lamb wave propagation group velocity and the frequency-thickness product of the present invention;
[0043] Figure 4 This is a flowchart of a Long Short-Term Memory (LSTM) network model in a specific embodiment of the present invention;
[0044] Figure 5 This is a schematic diagram of the random label matrix of the simulation loss in a specific embodiment of the present invention;
[0045] Figure 6 This is a specific embodiment of the present invention and Figure 5 The localization matrix diagram output by the corresponding Long Short-Term Memory (LSTM) network model;
[0046] Figure 7 This is a tag matrix diagram corresponding to the measured guided wave array signal of actual damage in a specific embodiment of the present invention;
[0047] Figure 8 This is a localization effect diagram of the first loss function for the measured guided wave array signal in a specific embodiment of the present invention;
[0048] Figure 9 This is a localization effect diagram of a second loss function with sparse weights introduced in a specific embodiment of the present invention. Detailed Implementation
[0049] The present application will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.
[0050] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.
[0051] Figure 1 The present invention illustrates an ultrasonic guided wave damage localization method for resin-based carbon fiber composite plates, the method comprising the following steps:
[0052] S1. Construct a sparse sensor waveguide matrix and obtain the waveguide array signal.
[0053] S11. Construct a sparse sensor waveguide matrix and determine the detection area: Based on W sensors, construct a sparse sensor waveguide matrix and set the area covered by the sparse sensor waveguide matrix as the detection area; the W sensors are numbered 1 to W respectively.
[0054] S12. Acquire several sets of transmitting and receiving sensor groups and waveguide array signals: sequentially excite sensors 1 to W as excitation sensors. During each excitation, W-1 sensors other than the excited sensor are used as receiving sensors, and the waveguide signals of all receiving sensors are collected at the same time. A total of W(W-1) sets of transmitting and receiving sensor groups and corresponding waveguide signals are obtained to form the waveguide array signal. Each transmitting and receiving sensor group includes 1 excitation sensor and 1 receiving sensor. The waveguide array signal includes W(W-1) sets of waveguide signals.
[0055] If sensor 3 transmits and sensor 1 receives, the propagation distance is the same as the distance from sensor 1 transmitting and sensor 3 receiving, and the Lamb wave waveform is therefore the same.
[0056] S13. Exclude waveguide signals with the same waveform from the waveguide array signals: In all transmitting and receiving sensor groups, only one sensor is retained if the excitation sensor and the receiving sensor are interchanged. At the same time, only one corresponding waveguide array signal is retained in the array signal. The waveguide array signals after exclusion include W(W-1) / 2 groups of waveguide signals.
[0057] In one specific embodiment, such as Figure 2As shown, a sparse sensor waveguide matrix consisting of W = 8 sensors forms a square detection area of 280mm × 280mm. Sensors 1-8 are excited sequentially, and the other 7 sensors simultaneously acquire waveguide signals during each excitation, resulting in a total of 56 sets of waveguide signals as the waveguide array signal. Signals with the same waveform are excluded from the waveguide array signal, resulting in a waveguide array signal containing 28 sets of waveguide signals.
[0058] S2. Establish a damage detection simulation model and obtain the corresponding damage detection training set and damage detection test set under simulated damage.
[0059] S21. Establish a damage detection simulation model: Divide the detection area into an N×N grid and construct an N×N label matrix corresponding to the simulated damage; if there is damage in the grid, the corresponding position of the label matrix is represented as 1; otherwise, the corresponding position of the label matrix is represented as 0; if there is damage in only one grid in the detection area, an N×N label matrix containing one 1 and (N×N-1) 0s is obtained.
[0060] S22. Calculate the distance of simulated damage propagation: Calculate the sum of the distances from the simulated damage to the excitation sensor and the receiving sensor, which is the distance the Lamb wave propagates from the excitation sensor to the damage reflection and then to the receiving sensor.
[0061] S23. Using the short pure tone toneburst signal g(t) as the excitation of the sensor, its expression is:
[0062]
[0063] Where f represents the center frequency; cyc represents the number of peaks; and t represents time.
[0064] S24. Perform a Fast Fourier Transform on g(t), and set the sampling frequency f. s M waveguide array signal samples were obtained.
[0065] S25. Obtain the waveguide array signal corresponding to the simulated damage: based on the Lamb wave propagation group velocity versus frequency-thickness product curve, such as... Figure 3 As shown, the propagation speed of guided wave signals in the plate at different center frequencies f is obtained, the frequency domain signal of the guided wave signal when the propagation distance is D is calculated, and the inverse Fourier transform and normalization of the guided wave signal are performed in sequence. White noise with an amplitude of r is added to obtain the guided wave array signal corresponding to the simulated damage.
[0066] S26. Construct damage detection training sets and damage detection test sets: Each simulated damage location corresponds to a set of guided wave array signals and a label matrix. Randomly select a points in each grid to represent simulated damage, and obtain a×N×N damage detection training sets. Randomly select b points in the entire detection area to obtain b damage detection test sets.
[0067] In one specific embodiment, such as Figure 2 As shown, the detection area is divided into a 20×20 grid, resulting in a 20×20 label matrix. The center frequency f of the short pure tone toneburst signal is 40000, and the number of peaks cyc = 3. The frequency domain signal is calculated when the guided wave signal propagates over a distance D. After normalization, white noise with an amplitude of 0.3 is added. Twenty points are randomly selected from each grid to represent damage, resulting in a training set of 8000 damage detection points. Thirty hundred damage points are randomly selected from the detection area as the test set for damage detection.
[0068] S3. Construct a Long Short-Term Memory (LSTM) network model: The LSTM network model includes LSTM layers and fully connected layers; preferably, there are 2 LSTM layers and 2 fully connected layers.
[0069] The output of the Long Short-Term Memory (LSTM) network model is the localization matrix; the first loss function of the LSTM network model is set as follows:
[0070] loss = (loc - label) 2 (2)
[0071] Where label represents the label matrix; loc represents the localization matrix output by the Long Short-Term Memory (LSTM) model in each training cycle; the localization matrix has the same size as the label matrix.
[0072] In one specific embodiment, Figure 4 The flowchart of the Long Short-Term Memory (LSTM) network model is shown. Figure 5 This diagram illustrates the random label matrix of the simulation loss. Figure 6 Showing with Figure 5 The corresponding localization matrix diagram output by the Long Short-Term Memory (LSTM) network model shows that the proposed LSTM network model has accurate localization capability and effect for the lost simulated waveguide array signal.
[0073] S4. Introduce sparse weights to obtain the second loss function.
[0074] S41. Obtain the localization matrix loc output by the Long Short-Term Memory (LSTM) network model in each training cycle.
[0075] S42. Introduce sparse weights and define a second loss function. x Represented as:
[0076] loss x =loss+λ|loc| (3)
[0077] Where λ represents sparse weights; |loc| represents the modulus of the localization matrix loc output by the LSTM model in each training cycle.
[0078] S5. Based on the damage detection training set and damage detection test set, train the Long Short-Term Memory (LSTM) network model and obtain the optimal LSTM network model.
[0079] The training hyperparameters are set as follows: 40 iterations, 64 samples per batch, 0.0005 learning rate, Adam optimizer, and 800 hidden layer dimension.
[0080] S6. Obtain damage localization imaging results based on measured guided wave array signals of actual damage: Acquire measured guided wave array signals of actual damage, input them into the optimal long short-term memory (LSTM) network model, and use the second loss function... x The output is the localization matrix, which is the damage localization imaging result.
[0081] In one specific embodiment, λ is set to 0.01. Figure 7 The diagram shows the tag matrix corresponding to the measured guided wave array signal of actual damage. Figure 8 The diagram shows the localization effect of the first loss function on the measured guided wave array signal. Figure 9 The localization effect of the second loss function with sparse weights is shown in the image. It can be seen that compared to... Figure 8 , Figure 9 This significantly reduces positioning noise.
[0082] This invention presents an ultrasonic guided wave damage localization method for resin-based carbon fiber composite plates. A sparse sensor guided wave matrix is constructed, and the simulated damage guided wave array signal is obtained using Fourier transform. Simulated damage detection experiments are then conducted. A Long Short-Term Memory (LSTM) network model is constructed using LSTM, successfully locating the damage position. By introducing sparse weights, the loss function of the LSTM model is improved, enhancing the accuracy of damage localization while reducing localization noise.
[0083] Finally, it should be noted that the above embodiments are for illustration only and not for limiting the technical solutions of the present invention. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the present invention without departing from the spirit and scope of the present invention. Any modifications or partial substitutions should be covered within the scope of the claims of the present invention.
Claims
1. A method for ultrasonic guided wave damage localization in resin based carbon fiber composite panels, characterized in that, It includes the following steps: S1. Construct a sparse sensor waveguide matrix and obtain the waveguide array signal; S2. Establish a damage detection simulation model and obtain the corresponding damage detection training set and damage detection test set under simulated damage. Step S2 specifically includes the following steps: S21. Establish a damage detection simulation model: Divide the detection area into an N×N grid and construct an N×N label matrix corresponding to the simulated damage. S22. Calculate the distance of simulated damage propagation: Calculate the sum of the distances from the simulated damage to the excitation sensor and the receiving sensor, which is the distance the Lamb wave propagates from the excitation sensor to the damage reflection and then to the receiving sensor. S23. The short pure tone toneburst signal g(t) is used as the excitation of the excitation sensor, and its expression is: (1) ； Where f represents the center frequency; cyc represents the number of peaks; and t represents time. S24, performing fast Fourier transform on g(t), setting sampling frequency f s obtain M samples of the guided wave array signal; S25. Obtain the waveguide array signal corresponding to the simulated damage: Based on the Lamb wave propagation group velocity and frequency-thickness product relationship curve, obtain the propagation velocity of the waveguide signal in the plate at different center frequencies f, calculate the frequency domain signal of the waveguide signal when the propagation distance is D, perform inverse Fourier transform and normalization of the waveguide signal in sequence, and add white noise with an amplitude of r to obtain the waveguide array signal corresponding to the simulated damage. S26. Construct the damage detection training set and damage detection test set: Each simulated damage location corresponds to a set of guided wave array signals and a label matrix. In each grid, a points are randomly selected to represent simulated damage, resulting in a × N × N damage detection training sets. In the entire detection area, b points are randomly selected to obtain b damage detection test sets. S3. Construct a Long Short-Term Memory (LSTM) network model: The LSTM network model includes LSTM layers and fully connected layers; the output of the LSTM network model is a localization matrix; the first loss function of the LSTM network model is set as follows: (2) ； Where label represents the label matrix; loc represents the localization matrix output by the Long Short-Term Memory (LSTM) network model in each training cycle; the localization matrix has the same size as the label matrix; S4. Introduce sparse weights to obtain the second loss function; S41. Obtain the localization matrix loc output by the Long Short-Term Memory (LSTM) network model in each training cycle; S42, introduce sparse weight, set the second loss function is represented as: (3) ； wherein, denotes a sparse weight; denotes a modulus of the localization matrix loc output by the long short-term memory network LSTM model in each training cycle; S5, based on the damage detection training set and the damage detection test set, with the aid of a second loss function training the long short-term memory network LSTM model to obtain an optimal long short-term memory network LSTM model; S6. Obtain damage localization imaging results based on measured guided wave array signals of actual damage: Collect measured guided wave array signals of actual damage, input the optimal Long Short-Term Memory (LSTM) network model, and output the localization matrix, i.e., the damage localization imaging results.
2. The ultrasonic guided wave damage localization method for resin-based carbon fiber composite plates according to claim 1, characterized in that, Step S1 specifically includes the following steps: S11. Construct a sparse sensor waveguide matrix and determine the detection area: Based on W sensors, construct a sparse sensor waveguide matrix and set the area covered by the sparse sensor waveguide matrix as the detection area; the W sensors are numbered 1 to W respectively; S12. Acquire several sets of transmitting and receiving sensor groups and waveguide array signals: sequentially excite the sensors numbered 1 to W as excitation sensors. Each time the sensors are excited, W-1 sensors other than the excited sensor are used as receiving sensors, and the waveguide signals of all the receiving sensors are collected at the same time. A total of W(W-1) sets of transmitting and receiving sensor groups and the corresponding waveguide signals are obtained to form the waveguide array signal. S13. Exclude waveguide signals with the same waveform from the waveguide array signals: In all the transmitting and receiving sensor groups, only one of the excitation sensor and the receiving sensor corresponding to the sensor is retained if they are interchanged. At the same time, only one of the corresponding waveguide array signals in the array signals is also retained.
3. The ultrasonic guided wave damage localization method for resin-based carbon fiber composite plates according to claim 2, characterized in that, In step S12, each of the transmitting and receiving sensor groups includes one excitation sensor and one receiving sensor; the guided wave array signal includes W(W-1) groups of guided wave signals; in step S13, the guided wave array signal after exclusion includes W(W-1) / 2 groups of guided wave signals.
4. The ultrasonic guided wave damage localization method for resin-based carbon fiber composite plates according to claim 1, characterized in that, In step S21, if there is damage within the grid, the corresponding position of the label matrix is represented as 1; otherwise, the corresponding position of the label matrix is represented as 0.
5. The ultrasonic guided wave damage localization method for resin-based carbon fiber composite plates according to claim 1, characterized in that, In step S3, there are two LSTM layers in the network; there are two fully connected layers.
6. The ultrasonic guided wave damage localization method for resin-based carbon fiber composite plates according to claim 1, characterized in that, In step S5, the hyperparameters used for training the Long Short-Term Memory (LSTM) network model are set as follows: the optimizer is Adam and the hidden layer dimension is 800.
7. The ultrasonic guided wave damage localization method for resin-based carbon fiber composite plates according to claim 1, characterized in that, In step S25, the amplitude r of the white noise is set to 0.3.