Composite material flat plate structure impact energy identification method based on impact stress wave and deep learning

By pasting piezoelectric sensors on the composite material and using deep learning network to identify impact stress wave signals, the problem that composite materials are difficult to monitor damage in real time under impact is solved, and efficient identification of impact position and energy is achieved, supporting the safety of the aircraft Assessment and repair guidance.

CN119025896APending Publication Date: 2024-11-26DALIAN UNIV OF TECH +1
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
CN202411226368.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-09-02
Publication Date
2024-11-26

AI Technical Summary

Technical Problem

During service, it is difficult to monitor impact damage online in real time, especially internal damage caused by low-speed impact. The existing non-destructive testing methods are time-consuming and labor-intensive, difficult to meet the needs of engineering applications, and there are few researches on impact energy recognition based on deep learning.

Method used

Using impact energy recognition method of composite plate structure based on impact stress waves and deep learning, ball-falling impact experiments are carried out by pasting piezoelectric sensors, impact guide signals are collected in real time, databases and deep learning network models are established, and time convolution networks and lengths are used. The memory neural network module recognizes impact location and energy.

Benefits of technology

It realizes accurate identification of impact position and energy without considering the wave reach time and peak value of impact stress wave signal, improves the efficiency and accuracy of impact energy recognition, provides real-time online monitoring capabilities, and supports aircraft safety evaluation and maintenance guidance. .

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119025896A_ABST
    Figure CN119025896A_ABST
Patent Text Reader

Abstract

The invention discloses a composite material flat plate structure impact energy identification method based on impact stress waves and deep learning, and belongs to the technical field of structure health monitoring. The method mainly comprises the steps of establishing a first-stage deep learning model for learning sequence features and high-dimensional abstract features of impact stress wave signals, and finally establishing a mapping relation between the impact stress wave signals and impact position coordinates; and establishing a second deep learning model for establishing a mapping relationship between the impact stress wave signal and the impact wave energy. According to the method provided by the invention, more impact stress wave signal characteristics can be learned, and it is indicated that the composite material impact position and energy identification technology combining deep learning and impact experiments has an application prospect.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of structural health monitoring, and in particular to an impact energy identification method for a composite material flat plate structure based on impact stress waves and deep learning. Background Art

[0002] Composite materials have high specific strength, high specific modulus, excellent fatigue resistance and corrosion resistance, and can effectively reduce the structural mass of aerospace vehicles. They are currently the focus of research in the field of new aerospace materials. However, composite materials have poor impact resistance and will inevitably suffer from external impacts during service. According to the size of the impact energy, they can be divided into high-speed impact and low-speed impact. High-speed impact will cause obvious penetrating damage to the surface of the structure, while low-speed impact will usually cause internal damage such as delamination and fiber breakage, which is not easy to observe directly with the naked eye. Therefore, it is difficult to judge whether the structure has been impacted and the degree of impact. If the damage is not discovered in time or the judgment is wrong, it will seriously threaten the service safety of the aircraft.

[0003] In response to the above problems, non-destructive testing technologies such as ultrasonic scanning, eddy current, and thermal imaging are currently commonly used to conduct regular inspections of impact damage. However, these methods are time-consuming and labor-intensive, and can only be used for offline detection, which is difficult to meet the needs of engineering applications. Impact monitoring technology can perform real-time online monitoring and has received widespread attention from scholars at home and abroad in recent years. This technology obtains real-time impact response signals through a sensor network and uses advanced signal processing methods and algorithms to identify the impact position and energy. Impact energy identification is one of the key technologies. It can estimate the impact energy, which is conducive to further analysis and judgment of the degree of damage, and provides a basis and guidance for subsequent offline detection and maintenance.

[0004] With the vigorous development of artificial intelligence technology, deep learning algorithms are increasingly used in the field of structural health monitoring. The impact energy identification method based on deep learning models has data mining capabilities independent of prior knowledge and can obtain a deep mapping relationship between high-dimensional data and target parameters. At present, there are relatively few studies on impact energy identification based on deep learning. Therefore, it is urgent to develop an efficient and reliable deep learning method to accurately identify the impact energy size. Summary of the invention

[0005] In view of the problems existing in the prior art, the present invention discloses a method for identifying impact energy of a composite flat plate structure based on impact stress waves and deep learning, and the steps are as follows: Paste a piezoelectric sensor and conduct an impact experiment. Paste a piezoelectric sensor on the composite structure to form a sensor network. Install four piezoelectric sensors on the test piece, conduct a falling ball impact experiment at different positions on the test piece in turn, and collect impact guided wave signals in real time; mark the data and establish a database. This step specifically includes: using the collected impact stress wave signal to mark with the impact position and energy labels to establish a database; establish a deep learning network model, including establishing a two-stage deep learning network to identify the impact position and energy; evaluate the model effect and position identification; and judge the effect of the model according to the evaluation index, and identify the impact point position coordinates and energy size.

[0006] In some embodiments, the stress wave response signal characteristic is a time series characteristic.

[0007] In some embodiments, an evaluation index is established to evaluate the effect of the deep learning network model, including using L1Loss as a loss function, calculating the absolute value of the prediction error of each sample, and taking the average value.

[0008] In some embodiments, using R 2 The score (R-squared) measures the degree to which the model explains the data variance, using the formula Calculate, where represents the mean of the target variable.

[0009] In some embodiments, the evaluation index determines the effect of the model and identifies the impact point location coordinates and energy magnitude.

[0010] Some other embodiments of the present application propose a method for identifying impact energy of a composite material flat plate structure based on impact stress wave and deep learning, which comprises the steps of: marking the collected impact stress wave signal, impact position and impact energy, and establishing a database; establishing a deep learning network model, including a time convolutional network module at the first level, a first convolutional neural network module, and a long short-term memory (LSTM) neural network module and a second convolutional neural network module at the second level; the deep learning network model is configured to use the time convolutional network module to identify the stress wave response signal feature from the impact stress wave signal; and using the first convolutional neural network to identify the stress wave response signal feature from the stress wave response. The invention discloses a method for realizing a mapping relationship between the impact stress wave signal and the impact position coordinates by learning a first high-dimensional abstract feature from the signal feature, and using a first fully-connected module with the first high-dimensional abstract feature as input and the impact source coordinates as output; using a long short-term memory (LSTM) neural network module to learn the energy feature representation of the impact stress wave signal from the stress wave response signal feature; and using a second convolutional neural network module to learn the second high-dimensional abstract feature of the stress wave response signal from the energy feature representation; and using a second fully-connected module to establish a mapping relationship between the impact stress wave signal and the impact source energy based on the second high-dimensional abstract feature.

[0011] In some embodiments, the stress wave response signal characteristic is a time series characteristic.

[0012] In some embodiments, an evaluation index is established to evaluate the effect of the deep learning network model, including using L1Loss as a loss function, calculating the absolute value of the prediction error of each sample, and taking the average value.

[0013] In some embodiments, using R 2 The score (R-squared) measures the degree to which the model explains the data variance, using the formula Calculate, where represents the mean of the target variable.

[0014] In some embodiments, the evaluation index determines the effect of the model and identifies the impact point location coordinates and energy magnitude.

[0015] Due to the adoption of the above-mentioned technical scheme, the advantage of the impact energy identification method of a composite material flat plate structure based on impact stress wave and deep learning provided by the present invention is that the impact position and energy are identified by using deep learning technology, without considering the arrival time, peak value and other information of the impact stress wave signal, and the deep-level characteristics of the data can be mined and learned. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0017] Figure 1 A schematic diagram of an impact experiment of a composite material flat plate structure impact energy identification method according to an embodiment of the present application;

[0018] Figure 2a , Figure 2b , Figure 2c and Figure 2d They are respectively schematic diagrams of impact stress wave signals of a composite material flat plate structure impact energy identification method according to an embodiment of the present application;

[0019] Figure 3a A schematic diagram of a deep learning network model of a composite material flat plate structure impact energy identification method according to an embodiment of the present application;

[0020] Figure 3b A schematic diagram of TCN dilated convolution in impact source position identification used in a method according to an embodiment of the present application;

[0021] Figure 3c A schematic diagram of the operating principle of a first CNN two-dimensional convolution module in a shock source position recognition model in a method according to an embodiment of the present application;

[0022] Figure 3d It is a schematic diagram of the principle of the LSTM module of the impact source energy identification model in the method according to the embodiment of the present application;

[0023] Figure 3e A schematic diagram of a deep learning network model of a composite material flat plate structure impact energy identification method according to an embodiment of the present application;

[0024] Figure 4 is a loss function iteration diagram of a composite material flat plate structure impact energy identification method according to an embodiment of the present application;

[0025] Figure 5a , Figure 5b , Figure 5c They are schematic diagrams of impact source positioning errors about points A, B and C in a composite material flat plate structure impact energy identification method according to an embodiment of the present application;

[0026] Figure 6It is a comparison chart of the actual value and the predicted value of the impact energy at points A, B and C according to the impact energy identification method of the composite material flat plate structure according to the embodiment of the present application. DETAILED DESCRIPTION

[0027] To make the technical solutions and advantages of the present invention more clear, the technical solutions in the embodiments of the present invention are clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention:

[0028] [Example 1: Impact energy identification]

[0029] A method for identifying impact damage energy of a composite plate structure based on impact stress wave and deep learning specifically comprises the following steps:

[0030] Paste the piezoelectric sensor and conduct a falling ball impact experiment.

[0031] A piezoelectric sensor PZT, such as a piezoelectric ceramic sensor, is pasted on four locations of the composite material structure. Figure 1 As shown. The falling ball impact experiment is carried out at different positions. For example, the falling ball impact is carried out at the position of the impact source SoC, and the shock wave signal propagates to the piezoelectric sensor along the signal propagation path SP. The piezoelectric sensor collects the impact waveguide signal and records the coordinate position of the impact point. The signal examples collected by the four sensors PZT1 to PZT4 are shown as follows: Figure 2a to Figure 2d As shown.

[0032] Labeling data and establishing a database includes labeling the collected impact stress wave signals, the position coordinates of the falling ball, and the impact energy and establishing a database.

[0033] Establish a deep learning network model. Specifically, it includes establishing a two-stage shock wave diagnosis method, wherein the first stage is a model that uses a time convolution module, i.e., a TCN module 110 to extract the time series characteristics of the shock response signal, i.e., the shock stress wave signal, and the second stage is a shock source energy identification model 200 established by using an LSTM-CNN network model to learn the magnitude of the shock energy. The overall network model structure of the method according to this embodiment is as follows: Figure 3a shown.

[0034] The collected impact stress wave signal is a complex sequence signal containing time information. The dimension of each data is recorded as C×H×W (1×4×5000), where C, H, and W represent the data channel, height, and width, respectively.

[0035] In the first stage, the impact stress wave signals input by each sensor are respectively passed through the time convolution network module, i.e., the TCN module 110 to obtain the time series characteristics of the impact stress wave signals. Specifically, the input impact stress wave signals are first passed through the time convolution network module 110 on each sensor signal to process and learn the time series characteristics of the impact stress wave signals, and then the signal time series characteristics of the four sensors are spliced, as shown in FIG. Figure 3b shown.

[0036] In the temporal convolutional network module 110, a dilated convolution calculation method can be used, which increases the size of the convolution kernel by inserting a fixed number of spaces between the elements of the standard convolution kernel without increasing the number of parameters or the amount of calculation. This calculation method can increase the receptive field of the convolution layer, that is, the size of the area covered by a single convolution operation in the input feature map, allowing the network to capture a wider range of contextual information without significantly increasing the computational burden. The calculation process is shown in formula (1).

[0037]

[0038] Among them, i is the position index in the output sequence, k is the size of the convolution kernel, w is the convolution kernel, x[i] is the input sequence, f[i] is the output sequence, l is the offset position index of the convolution kernel, indicating the lth position of the convolution kernel, and d is the dilation factor, which defines the step size of the convolution kernel jumping on the input sequence.

[0039] In the second level, i.e., the impact energy identification model 200, the stress wave response signal characteristics identified by the first-level temporal convolutional network module 110 are firstly used to learn the energy characteristic representation of the impact stress wave signal through the long short-term memory (LSTM) neural network module 210, and then the energy characteristic representation of the impact stress wave signal is learned through the second convolutional neural network module 220, i.e., the second CNN module 220, and the high-dimensional abstract characteristics of the signal are finally mapped with the impact source energy.

[0040] The high-dimensional abstract features reflect the spatial coupling characteristics of the stress wave response signals detected by the four piezoelectric sensors PZT1 to PZT3.

[0041] In this embodiment, the LSTM neural network module 210 may include two layers of long short-term memory (LSTM) units. The structure of each LSTM network unit is as follows: Figure 3d As shown, it includes a forget gate, an input gate, and an output gate. Therefore, the calculation process of the LSTM network unit is shown in formula (2):

[0042] f t =σ(W f ·[h t-1 ,x t ]+bf )

[0043] i t =σ(W i ·[h t-1 ,x t ]+b i )

[0044]

[0045] c t =f t *c t-1 +i t *c t

[0046] o t =σ(W o ·[h t-1 ,x t ]+b o )

[0047] h t =o t *tanh(c t ) (2)

[0048] Among them, x t is the input at the current moment, f t is the activation value of the forget gate, h t-1 is the hidden state at the previous moment, b f is the bias vector of the forget gate, W f is the weight matrix of the forget gate, σ is the sigmoid activation function; i t is the input gate; is the candidate memory unit value, tanh is the tanh activation function; c t is the memory cell value at the current moment, c t-1 is the memory cell value at the previous moment, * represents element multiplication (Hadamard product); o t is the output gate, h t is the hidden state at the current moment, t represents the current time step, and t-1 represents the previous time step.

[0049] The second CNN module 220 is as follows Figure 3c As shown, it includes two layers of convolution calculation layers. After three layers of convolution layers with a convolution kernel size of C×3×3 and a stride of 2×2, a maximum pooling operation of 2×2 and a stride of 2×2 is performed. The convolution process is to multiply the original data value by the corresponding convolution kernel value, and then sum it as the result of the convolution calculation. Then the convolution kernel slides on the original data with a certain step size and performs convolution calculation until the entire data slides. The calculation process of the two-dimensional convolution is shown in formula (3).

[0050]

[0051] Among them, k is the size of the convolution kernel, f(i,j) is the original data, and g(ki,kj) is the convolution kernel.

[0052] In each layer of the second CNN module 220, batch normalization (BN) and ReLU activation function processing are performed after each convolution calculation, thereby promoting the training efficiency and performance of the neural network. Batch normalization normalizes each small batch of data so that the distribution of each layer input remains relatively stable, reduces the dependence on parameter initialization during training, and reduces the sensitivity of the model to the initial weights. The normalization process of each batch adds a little noise, which has a regularization effect to a certain extent and helps prevent the model from overfitting. The ReLU activation function can increase the nonlinearity of the model, help alleviate the gradient vanishing problem in deep networks, and enable deeper networks to be successfully trained.

[0053] The calculation process of batch normalization is shown in formula (4):

[0054]

[0055] Among them, x i represents the i-th sample, m is the batch size, is the mean, is the variance, ε is a small constant used to prevent division by zero, and γ and β are learnable parameters for scaling and offset, respectively.

[0056] The data calculated by the second CNN module 220 is finally converted into a one-dimensional signal feature through a full connection operation of the full connection module 230, and finally a mapping relationship is established with the energy of the impact.

[0057] L1 Loss (also known as absolute error loss) is a loss function used to measure the difference between two vectors. i and The formula (5) of L1Loss is as follows:

[0058]

[0059] They represent the true value and corresponding predicted value of the i-th sample respectively, and N is the number of samples.

[0060] L1 Loss represents the sum of the absolute differences between the elements at corresponding positions in two vectors, so it is sensitive to outliers. L1 Loss is used as a loss function in many machine learning tasks, such as regression problems, where the goal is to minimize the absolute difference between the predicted value and the actual value.

[0061] In the process of deep learning back propagation, the optimizer is an optimization algorithm for solving the loss function, guiding the various parameters of the loss function to update the appropriate values ​​in the right direction, so that the parameter values ​​of the updated loss function are constantly approaching the global minimum. The RMSProp optimization method is adopted in the present invention, which is an adaptive learning rate optimization algorithm, which aims to solve the problem of learning rate adjustment in the gradient descent process. By performing an exponentially weighted moving average on the square of the gradient, the learning rate of each parameter is adjusted to ensure that the parameter update is stable and converges quickly. It can effectively prevent the oscillation phenomenon in the gradient descent process by performing an exponentially weighted moving average on the square of the gradient. The optimization algorithm first initializes the exponentially weighted moving average and time step of the square of the gradient, and then calculates the current gradient, and updates the exponentially weighted moving average and learning rate of the square of the gradient, as shown in formula (6).

[0062] E[g 2 ]0=0,t=0

[0063] E[g 2 ] t =βE[g 2 ] t-1 +(1-β)g t 2 (6)

[0064]

[0065] Where E[g 2 ] t is the exponentially weighted moving average of the squared gradient, g t is the current gradient, β is the decay rate, θ t is the parameter value at the current moment, η is the global learning rate, and ε is a small constant used for numerical stability. The recommended default value is 1e-8.

[0066] Evaluation indicators:

[0067] In regression tasks, the indicators for evaluating model performance can help understand the model's fit and prediction accuracy. 2 As an evaluation indicator. L1 Loss calculates the absolute value of the prediction error of each sample and takes the average value. The smaller the value, the better the model fit. 2The score (R-squared) is used to measure the degree to which the model explains the data variance. The closer it is to 1, the better the effect. The calculation method is shown in formula (7):

[0068]

[0069] in represents the mean of the target variable. R 2 A value equal to 1 indicates a perfect model fit.

[0070] [Example 2: Impact position + energy identification]

[0071] In other embodiments of the present application, the location of the impact source can be predicted based on the TCN-CNN network learning model 100, and then the signal time series features obtained by the TCN module 110 in the TCN-CNN network learning model 100 are spliced ​​and sent to the impact energy identification module 200 to identify the energy of the shock wave.

[0072] This embodiment 2 is similar to the embodiment 1, but the establishment of the deep learning network model specifically includes establishing a two-stage shock wave diagnosis method, wherein the first stage is a shock source position identification model 100 established by using a TCN-CNN (time convolution-convolution) network learning model to identify the location of the shock source, and the second stage is a shock source energy identification model 200 established by using an LSTM-CNN network model to learn the magnitude of the shock energy. The overall network model structure of the method according to the embodiment of the present application is as follows: Figure 3e shown.

[0073] In this embodiment, firstly, the collected impact stress wave signals, impact positions and impact energies are annotated to establish a database;

[0074] A two-level deep learning network model as described above is established, including a time convolutional network module 110 at the first level, a first convolutional neural network module 120, and a long short-term memory (LSTM) neural network module 210 and a second convolutional neural network module 220 at the second level; the deep learning network model is configured to use the time convolutional network module 110 to identify the stress wave response signal feature from the impact stress wave signal; use the first convolutional neural network 120 to learn the first high-dimensional abstract feature from the stress wave response signal feature, and use the first fully connected module 130 to use the first high-dimensional abstract feature as input and the impact source coordinates as output to establish a mapping relationship between the impact stress wave signal and the impact position coordinates; use the long short-term memory (LSTM) neural network module 210 to learn the energy feature representation of the impact stress wave signal from the stress wave response signal feature; and use the second convolutional neural network module 220 to learn the second high-dimensional abstract feature of the stress wave response signal from the energy feature representation; and use the second fully connected module 230 to establish a mapping relationship between the impact stress wave signal and the impact source energy based on the second high-dimensional abstract feature.

[0075] The first high-dimensional abstract feature and the second high-dimensional abstract feature reflect the spatial coupling characteristics of the stress wave response signals detected by the four piezoelectric sensors PZT1 to PZT3.

[0076] Evaluation indicators:

[0077] In regression tasks, the indicators for evaluating model performance can help understand the model's fit and prediction accuracy. 2 As an evaluation indicator. L1 Loss calculates the absolute value of the prediction error of each sample and takes the average value. The smaller the value, the better the model fit. 2 The score (R-squared) is used to measure the degree to which the model explains the data variance. The closer it is to 1, the better the effect. The calculation method is shown in formula (7):

[0078]

[0079] in represents the mean of the target variable. R 2 A value equal to 1 indicates a perfect model fit.

[0080] The distance error between the actual position of the impact point and the predicted result is used as an intuitive evaluation criterion. The error formula is shown in formula (8):

[0081]

[0082] Where (x i ,yi ) represents the coordinates of the impact point predicted by the model, Represents the real impact source coordinates.

[0083] Experimental example:

[0084] To verify the effectiveness of the method of the present invention, Figure 1 As shown in Figure 2, a drop ball impact test was carried out on a carbon fiber composite reinforced wall panel with an overall size of 500mm×500mm. Four piezoelectric sensors, PZT1, PZT2, PZT3 and PZT4, were arranged at the four corners 70mm away from the boundary, and the sensor spacing was 360mm×360mm. Drop ball impact tests were carried out at different positions within the sensor envelope area, and impact stress wave signals were collected in real time, as shown in Figure 2, which shows the signals measured by the four sensors. Taking the upper left corner of the test piece as the origin, using impact energies of 0.1, 0.2, 0.3, 0.4, and 0.5J, a total of 1745 impact stress wave signals were collected.

[0085] The algorithm of the present invention was implemented through PyTorch induction experiment and two NVIDIAGeForce RTX3090s. Adam was used to optimize the model parameters, the weight decay was 1e-5, and the initial learning rate was 1e-4. The cosine annealing algorithm was used to gradually reduce the learning rate. The ratio of the training set and the test set was 8:2, the batch size was set to 32, and a total of 100 epochs were iterated. The loss curve is shown in the figure below. Figure 4 The impact position and energy recognition comparison results are shown in Figures 5 and 6, where the black solid points represent the real impact points and the square points represent the model prediction results. After multiple iterations of training, the R 2 is 0.9815, and in the impact energy identification task, it is R 2 is 0.9829, indicating that the model has a good fitting effect. Figure 5a , Figure 5b , Figure 5c The actual values ​​and predicted values ​​of the positions of the three points A, B, and C are represented by dots and squares respectively, as shown in Figure 5a As shown in Figure 2, the error between the predicted value and the true value of point A is 20.62 mm. Figure 5b As shown, the error of point B is 13.42mm; Figure 5c As shown, the error of point C is 6.40mm. Figure 6 As shown in, it is the true value and predicted value of the energy of the three points A, B, and C, such as Figure 6 As shown, the predicted value of point A is 0.52J, and the true value is 0.5J; the predicted value of point B is 0.08J, and the true value is 0.1J; the predicted value of point C is 0.35J, and the true value is 0.3J.

[0086] The results show that the deep learning algorithm can be used to learn the relationship between the impact stress wave signal and the impact position and energy without analyzing the signal characteristics. The deep learning method for identifying the impact position and energy of composite materials has broad application prospects.

[0087] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for identifying impact energy of composite plate structures based on impact stress waves and deep learning, characterized in that: Includes steps: Label the collected impact stress wave signals and impact energy and establish a database; Establish a deep learning network model, including a temporal convolutional network module at the first level, and a long short-term memory (LSTM) neural network module and a convolutional neural network module at the second level; Wherein, the deep learning network model is configured to utilize the stress wave response signal features identified from the impact stress wave signal by the time convolution network module; Using a long short-term memory (LSTM) neural network module to learn the energy feature representation of the stress wave response signal feature; and A convolutional neural network module is used to learn the high-dimensional abstract features represented by the energy characteristics of the impact stress wave signal, so as to establish a mapping relationship between the impact stress wave signal and the impact energy.

2. The method for identifying impact energy of composite material flat plate structure based on impact stress wave and deep learning according to claim 1, characterized in that: The stress wave response signal characteristic is a time series characteristic.

3. The impact energy identification method of composite material flat plate structure based on impact stress wave and deep learning according to claim 1, characterized in that: Evaluation indicators were established to evaluate the effect of the deep learning network model, including using L1Loss as a loss function, calculating the absolute value of the prediction error of each sample, and taking the average value.

4. The method for identifying impact energy of composite material flat plate structure based on impact stress wave and deep learning according to claim 1, characterized in that: Using R 2 The score (R-squared) measures the degree to which the model explains the data variance, using the formula Calculate, where represents the mean of the target variable.

5. The impact energy identification method of composite material flat plate structure based on impact stress wave and deep learning according to claim 1, characterized in that: The evaluation index determines the effect of the model and identifies the impact point location coordinates and energy size.

6. A composite material flat plate structure impact energy identification method based on impact stress wave and deep learning, characterized in that: Includes steps: Mark the collected impact stress wave signals, impact positions and impact energies, and establish a database; Establishing a deep learning network model, including a temporal convolutional network module at a first level, a first convolutional neural network module, and a long short-term memory (LSTM) neural network module and a second convolutional neural network module at a second level; The deep learning network model is configured to use a time convolutional network module to identify stress wave response signal features from the impact stress wave signal; use the first convolutional neural network to learn a first high-dimensional abstract feature from the stress wave response signal feature; and use a first fully connected module to use the first high-dimensional abstract feature as input and the impact source coordinates as output to establish a mapping relationship between the impact stress wave signal and the impact position coordinates; Using a long short-term memory (LSTM) neural network module to learn the energy feature representation of the impact stress wave signal from the stress wave response signal feature; and using a second convolutional neural network module to learn the second high-dimensional abstract feature of the stress wave response signal from the energy feature representation; and A mapping relationship between the impact stress wave signal and the impact source energy is established based on the second high-dimensional abstract feature using a second fully connected module.

7. The method for identifying impact energy of composite material flat plate structure based on impact stress wave and deep learning according to claim 1, characterized in that: The stress wave response signal characteristic is a time series characteristic.

8. The method for identifying impact energy of composite material flat plate structure based on impact stress wave and deep learning according to claim 1, characterized in that: Evaluation indicators were established to evaluate the effect of the deep learning network model, including using L1Loss as a loss function, calculating the absolute value of the prediction error of each sample, and taking the average value.

9. The impact energy identification method of composite material flat plate structure based on impact stress wave and deep learning according to claim 1, characterized in that: Using R 2 The score (R-squared) measures the degree to which the model explains the data variance, using the formula Calculate, where represents the mean of the target variable.

10. The method for identifying impact energy of composite material flat plate structure based on impact stress wave and deep learning according to claim 1, characterized in that: The evaluation index determines the effect of the model and identifies the impact point location coordinates and energy size.

Citation Information

Patent Citations

  • Variable-thickness composite material laminated plate low-speed impact energy identification system and operating method thereof

    CN108051126A

  • Impact localization method based on phase-sensitive optical reflection and deep learning of convolutional neural network

    CN108645498A

  • Estimating method for impact position and mass of metal piece using ambiguity pattern recognition

    KR101724151B1