Fatigue crack propagation and residual fatigue life prediction method fusing linear and nonlinear ultrasonic characteristics
By fusing linear and nonlinear ultrasound features and using LSTM neural networks, the fatigue crack propagation and residual fatigue life of the board are predicted, and the problem of difficulty in covering the contradiction between the entire life cycle and data volume in the prior art is solved, achieving high-precision and generalization prediction effects.
Patent Information
- Application Number
- CN202411867179.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-18
- Publication Date
- 2025-04-11
AI Technical Summary
The prior art is difficult to cover the entire life cycle of the material in fatigue crack propagation and residual fatigue life prediction, and it is difficult to balance the training data and historical information amount.
Using a method of fusing linear and nonlinear ultrasonic features, the ultrasonic guided wave signal under periodic loading force is obtained, linear and nonlinear ultrasonic characteristic parameters are calculated, and the weight is calculated based on the number of periods of the sampling points, and weighting is performed. Then, using the trained LSTM neural network, the crack length and/or remaining fatigue life of the plate to be tested are predicted.
It realizes high-precision prediction of the entire life cycle of the material, improves prediction accuracy, and has a certain generalization, and can adapt to loading capacity changes. At the same time, through the first set of data expansion method, the amount of data is increased, and the contradiction between training data and historical information is solved.
Smart Images
Figure CN119804673A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of fatigue crack propagation and fatigue life prediction of plate-like structures, and in particular to a fatigue crack propagation and residual fatigue life prediction method integrating linear and nonlinear ultrasonic features. Background Art
[0002] Fatigue refers to the gradual cracking or sudden fracture of materials under cyclic loads, and is the main cause of failure of metal structures. Fatigue is the main cause of failure of metal structures and may lead to serious accidents. Accidents. Studies have shown that 90% of mechanical structure failures are caused by fatigue cracks. Therefore, it is of great significance to perform fatigue crack detection on structures to ensure the safety of structures. However, routine maintenance based solely on detection is both inefficient and resource-intensive. Therefore, predicting fatigue crack growth and remaining fatigue life (RFL) is crucial for predictive maintenance. Traditional prediction methods based on physical models have limited effect on achieving accurate prediction of cracks due to the non-fidelity of their models.
[0003] In recent years, data-driven methods have gradually been applied in fatigue crack propagation and life prediction due to their advantages in processing high-dimensional features. Most researchers use linear ultrasonic features or nonlinear ultrasonic features for prediction. Among them, linear ultrasonic parameters can only detect macro cracks, while nonlinear ultrasonic parameters are required for monitoring the micro damage stage. In order to solve the problem that the prediction cannot cover the entire life cycle of the material, a method combining linear and nonlinear ultrasonic features is needed to achieve high-precision prediction over the entire life cycle.
[0004] There is a problem in using ultrasonic guided waves to obtain features and realize fatigue crack growth and life prediction based on data-driven methods: it is difficult to obtain sufficient data for neural network training, and sufficient data volume is the key to determining the prediction performance of neural networks. In addition, LSTM neural networks require sufficient historical information to learn on a time scale, and insufficient data often leads to a contradiction between the amount of training data and the amount of historical information of training data. Therefore, in order to deal with such engineering application prediction problems, it is necessary to solve the problem of training and historical information volume. Summary of the invention
[0005] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and to provide a fatigue crack propagation and residual fatigue life prediction method that integrates linear and nonlinear ultrasonic features, so as to solve or partially solve the problem of unsatisfactory accuracy of the results predicted based on a single feature.
[0006] The purpose of the present invention can be achieved by the following technical solutions:
[0007] One aspect of the present invention provides a fatigue crack growth and residual fatigue life prediction method integrating linear and nonlinear ultrasonic features, comprising:
[0008] Obtaining the ultrasonic guided wave signal of the plate to be tested under the periodic loading force, and calculating the linear ultrasonic characteristic parameters and nonlinear ultrasonic characteristic parameters of each sampling point;
[0009] Calculate the weight corresponding to the sampling point based on the cycle number where the sampling point is located, and weight the linear ultrasonic characteristic parameter and the nonlinear ultrasonic characteristic parameter based on the weight;
[0010] Based on the weighted linear ultrasonic characteristic parameters and nonlinear ultrasonic characteristic parameters of multiple sampling points, the crack length and / or residual fatigue life of the plate to be tested are predicted using the trained LSTM neural network.
[0011] As a preferred technical solution, the LSTM neural network learns the relationship between data through a sliding window, and after weighting the linear ultrasonic feature parameters and the nonlinear ultrasonic feature parameters, further includes:
[0012] Based on the weighted linear ultrasonic characteristic parameters and nonlinear ultrasonic characteristic parameters, the first set of data expansion is achieved by copying the first sampling points of multiple ultrasonic guided wave signals.
[0013] As a preferred technical solution, the calculation of the weights corresponding to the sampling points includes the following steps:
[0014] The fatigue stage is pre-divided into early, middle and late stages. In the early and middle stages, the weight of the nonlinear ultrasonic characteristic parameters is higher than that of the linear ultrasonic characteristic parameters. In the late stage, the weight of the linear ultrasonic characteristic parameters is higher than that of the early and middle stages.
[0015] As a preferred technical solution, a weighting function is used to calculate the weights corresponding to the sampling points. The weighting function is a concave function. In the early stage, the weight of the nonlinear ultrasonic characteristic parameter is greater than 0.9, and in the middle and late stages, the weight of the nonlinear ultrasonic characteristic parameter is greater than 0.7.
[0016] As a preferred technical solution, the weights corresponding to the sampling points are calculated using the following formula:
[0017]
[0018] Where N is the number of sampling points, N a is the average total life cycle of all plates, and δ(N) is the weight corresponding to the sampling point.
[0019] As a preferred technical solution, weighting the linear ultrasonic characteristic parameters and the nonlinear ultrasonic characteristic parameters based on the weights is implemented by the following formula:
[0020] β=β×δ(N)
[0021] A1=A1×(1-δ(N))
[0022] Among them, A1 and β are linear ultrasonic characteristic parameters and nonlinear ultrasonic characteristic parameters respectively, and δ(N) is the weight corresponding to the sampling point.
[0023] As a preferred technical solution, it includes two LSTM neural networks, namely a first LSTM neural network and a second LSTM neural network, wherein the inputs of the first LSTM neural network and the second LSTM neural network both include the number of fatigue cycles, cumulative characteristics, and weighted linear ultrasonic characteristic parameters and nonlinear ultrasonic characteristic parameters, the output of the first LSTM neural network is the predicted crack length of the plate to be tested, and the output of the second LSTM neural network is the predicted remaining fatigue life of the plate to be tested.
[0024] As a preferred technical solution, the calculation process of the linear ultrasonic characteristic parameters and the nonlinear ultrasonic characteristic parameters of the sampling points includes:
[0025] Based on the ultrasonic guided wave signal, the fundamental frequency amplitude A1 and the second harmonic amplitude A2 corresponding to the ultrasound are obtained by short-time Fourier transform, and the fundamental frequency amplitude A1 is used as the linear ultrasonic characteristic parameter. as a nonlinear characteristic parameter.
[0026] As a preferred technical solution, the LSTM neural network includes multiple LSTM layers and multiple fully connected layers connected in sequence, and for any fully connected layer, a dropout layer is included after the fully connected layer.
[0027] Another aspect of the present invention provides a fatigue crack growth and remaining fatigue life prediction system integrating linear and nonlinear ultrasonic features, which is used to implement the aforementioned fatigue crack growth and remaining fatigue life prediction method integrating linear and nonlinear ultrasonic features. The prediction system comprises:
[0028] An ultrasonic monitoring data acquisition module is used to obtain ultrasonic guided wave signals of the panel under periodic loading force, wherein the ultrasonic monitoring data acquisition module includes multiple groups of piezoelectric sheets, a part of which is used to excite ultrasonic guided waves, and another part of which is used to collect ultrasonic guided wave signals;
[0029] Linear and nonlinear feature extraction module, used to calculate the linear ultrasonic feature parameters and nonlinear ultrasonic feature parameters of each sampling point;
[0030] A feature fusion module, used to calculate the weight corresponding to the sampling point based on the number of cycles in which the sampling point is located, and weight the linear ultrasonic feature parameter and the nonlinear ultrasonic feature parameter based on the weight;
[0031] The model prediction module is used to predict the crack length and / or remaining fatigue life of the plate to be tested based on the weighted linear ultrasonic characteristic parameters and nonlinear ultrasonic characteristic parameters of multiple sampling points and using the trained LSTM neural network.
[0032] Compared with the prior art, the present invention has at least one of the following beneficial effects:
[0033] (1) Fusion of linear and nonlinear ultrasonic features for prediction: The present invention utilizes the linear and nonlinear fusion features of ultrasonic guided wave signals and realizes the prediction of fatigue crack propagation and residual fatigue life in plate structures based on machine learning.
[0034] (2) High prediction accuracy: Compared with the traditional single feature prediction, the prediction based on the fusion of linear and nonlinear ultrasonic features can cover the entire life cycle of the material and improve the prediction accuracy.
[0035] (3) The proposed prediction method has a certain degree of generalization and can adapt to the effective prediction under the loading force change of 5%;
[0036] (4) Fully consider the impact of historical data on prediction results: The present invention increases the amount of data by introducing the first group of data expansion method, and combines it with the LSTM network based on sliding window learning data connections, so that the neural network can make more use of historical data, which to a certain extent solves the contradiction between training data and historical information. BRIEF DESCRIPTION OF THE DRAWINGS
[0037] Figure 1 It is a flow chart of a fatigue crack growth and residual fatigue life prediction method integrating linear and nonlinear ultrasonic features in an embodiment;
[0038] Figure 2 A schematic diagram of a fatigue crack growth and residual fatigue life prediction method in an embodiment;
[0039] Figure 3 The dimensions of the aluminum plate and the arrangement of the PZTs used in the examples are shown below;
[0040] Figure 4 This is a schematic diagram of a weight function integrating linearity and nonlinearity in an embodiment;
[0041] Figure 5 It is a schematic diagram of fatigue crack and life prediction results in the embodiment;
[0042] Figure 6 This is a comparison diagram of the ultrasonic signal and the predicted measurement under the 5% change condition in the embodiment;
[0043] Figure 7 This is a schematic diagram of the prediction results under a 5% change in working condition fluctuation in the embodiment;
[0044] Figure 8 It is a schematic diagram of a fatigue crack growth and residual fatigue life prediction system integrating linear and nonlinear ultrasonic features in an embodiment;
[0045] Fig. 9 Schematic diagram of an electronic device in an embodiment. DETAILED DESCRIPTION
[0046] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.
[0047] Example 1
[0048] In view of the problems existing in the above-mentioned prior art, this embodiment provides a fatigue crack growth and residual fatigue life prediction method based on machine learning and integrating linear and nonlinear ultrasonic features. Figure 2 This method dynamically adjusts the proportion of the two features through a weighting function, inputs linear and nonlinear feature parameters into the LSTM model, and uses crack length or remaining life as output to achieve prediction of the entire life of the structure. The main purpose of this method is to overcome the problem that the prediction of fatigue crack growth and life of plate-like structures cannot cover the entire life cycle and it is difficult to balance the amount of training and historical information. This method provides a method for predicting fatigue crack growth and life by guided ultrasonic guided waves that integrates linear and nonlinear ultrasonic features and machine learning. This method can reasonably combine linear and nonlinear ultrasonic features and simply and conveniently expand the data set. Based on the establishment of a neural network model, the prediction of fatigue crack growth and life of plate-like structures can be achieved.
[0049] See also Figure 1 , the method comprises the following steps:
[0050] Step S1, ultrasonic monitoring and data acquisition: perform periodic fatigue loading on the metal plate structure, use ultrasonic guided waves to detect, and collect ultrasonic guided wave signals
[0051] Specifically, in multiple groups of metal plate specimens (Al6061, 300×200×2mm 3 ) is subjected to periodic loading (5 Hz, stress ratio 0.1), and multiple groups of piezoelectric transducers (PZTs) are used to obtain the corresponding signals of ultrasonic guided waves under different loading cycles, such as Figure 3 As shown, 1, 2, 3, and 4 in the figure represent four groups of piezoelectric sheets, 12 is used for excitation and 34 is used for receiving. At the same time, after each group of measurements, a CCD high-speed camera is connected to the PC software to take crack photos and obtain the crack length.
[0052] Step S2, linear and feature extraction: Calculate linear and nonlinear ultrasonic feature parameters based on the ultrasonic guided wave signal.
[0053] Specifically, the short-time Fourier transform (STFT) is used to obtain the fundamental frequency amplitude A1 and the second harmonic amplitude A2 of each group of ultrasound, and A1 is used as the linear feature. As a nonlinear feature, the characteristic parameters of each sampling point are obtained in this way.
[0054] Step S3: feature fusion.
[0055] Linear and nonlinear characteristics are respectively suitable for monitoring different fatigue stages. Nonlinear characteristics have higher weights in the early and middle stages, while linear characteristics have higher weights in the later stages. The division of fatigue stages is based on engineering experience. The first 70% of life is defined as the early and middle stages of fatigue, and the last 30% is defined as the late stage of fatigue. The weighting function is a concave function, which makes the weight of nonlinear characteristics reach above 0.9 in the early stage, and can still maintain above 0.7 in the middle and late stages, and drops sharply in the later stage, while linear characteristics rise rapidly in the later stage, and the weight change curve conforms to the law of fatigue evolution.
[0056] Specifically, the fusion of ultrasonic linear and nonlinear features uses a concave weighting function to combine linear and nonlinear features, so that the nonlinear features have a higher weight in the early stage, while the linear features gradually dominate in the later stage, such as Figure 4 As shown, it conforms to the characteristic variation law of structural fatigue evolution. The specific expression of the weight function is:
[0057]
[0058] Where N is the number of sampling points, N a is the average total life cycle of all panels. N a This setting is taken into consideration to prevent the problem that the test set cannot obtain the total number of cycles.
[0059] After obtaining the weight function, multiply it with the linear and nonlinear features respectively to obtain the weighted feature value as part of the subsequent neural network input:
[0060] β=β×δ(N)
[0061] A1=A1×(1-δ(N))
[0062] Step S4, LSTM model prediction. Based on the LSTM neural network model, the number of fatigue cycles, linear features, nonlinear features and their cumulative features BI are used as inputs, and the crack length or remaining fatigue life is used as output. The neural network model is trained to accurately predict crack extension and remaining fatigue life. Among them, the calculation method of BI is that when β increases, BI increases, and when β decreases, BI remains unchanged, as shown in the following formula:
[0063]
[0064] (1) Model construction.
[0065] Two LSTM neural networks are constructed respectively, whose inputs are the number of fatigue cycles, linear features, nonlinear features and cumulative features, and whose outputs are crack length and remaining fatigue life respectively. The two neural networks are used to fit the two parameters to be predicted respectively to make the prediction results more accurate.
[0066] Specifically, the LSTM neural network model includes two LSTM layers and two fully connected layers, and learns the connections between data through sliding windows to obtain more complex features.
[0067] The LSTM neural network of this model is constructed with the number of fatigue cycles, linear features, nonlinear features and cumulative features as inputs, and the crack length or remaining fatigue life as output. Two LSTM layers are set to obtain more historical information, and two fully connected layers are set after LSTM. There is a dropout layer after each fully connected layer, and the dropout rate is set to 0.2. Finally, the output of the quantity to be predicted is obtained. Sigmoid is selected as the nonlinear activation function in this neural network, and the Adam optimization algorithm is used for learning. The learning rate is set to 0.001, the loss function is selected as MSE, and the number of iterations is set to 450.
[0068] (2) Expansion of the first set of data.
[0069] In the training process of the LSTM neural network model, the padding method is used to prevent the loss of image edge information. Inspired by this, this embodiment provides a first group data expansion method. This method expands the data by copying a certain number of first sampling points. This is experimentally explainable because the copied data of the first sampling point is sampled before the experimental equipment is started. It can be considered that a certain amount of data has been obtained before the experiment starts.
[0070] The purpose of this is to enable the sliding window to cover all the original data so that the LSTM layer can learn more historical data rather than partial data. Another benefit of this method is that since the sliding window starts from the first sampling point, the data used for training will increase. In traditional LSTM networks, the sliding window calculates its size from the sampling point.
[0071] The first data augmentation method is used to add several sets of data at the front end of the data set to expand the data set, and all its eigenvalues are equal to the eigenvalues of the first set of collected data. After using the first data augmentation method, the LSTM neural network sliding window can be extended to before the first sampling point, so that the neural network can include more historical information in training and have more training sets.
[0072] A multidimensional feature data set is constructed by combining linear and nonlinear ultrasonic features using a weighted function, and the amount of the data set is increased using the first set of data augmentation method. The increased data set is physically interpretable and can be regarded as multiple collections of various features in the initial state.
[0073] (3) Dataset division.
[0074] In order to improve the accuracy of the model and the convergence speed of model training, the data is standardized and converted into dimensionless values that conform to the normal distribution of N(0,1). In addition, because the data fitting ability of the neural network is too strong, if samples with the same arrangement pattern appear repeatedly in a batch, the neural network will learn the arrangement pattern as a feature, which is prone to overfitting and reduces the generalization ability of the model. To address the above problems, the order of samples in the data set is disrupted before training. Finally, the data set is divided into two parts: 66% is the training set and 33% is the test set. In addition, 33% of the training set is used as a validation set.
[0075] (4) Model verification and evaluation.
[0076] The model was validated using a test set, and the results showed that the prediction effect was significant under different loading conditions and it had accurate prediction capabilities, such as Figure 5 As shown, the first line is the crack prediction results of test sets 1-3, and the second line is the remaining fatigue life prediction results of test sets 1-3.
[0077] In order to verify the generalization of the model, the data set under the condition of 5% load change was predicted. Although the change range was only 5%, the corresponding ultrasonic features and the predicted values had significant changes, such as Figure 6 As shown in the figure, the first line shows the curves of linear and nonlinear characteristics changing with loading cycles under three working conditions, and the second line shows the curves of crack length and residual fatigue life changing with loading cycles under three working conditions. The model is used to predict the data under unknown loading force, and the following is obtained: Figure 7 The results shown in the figure are as follows: the first row is the crack prediction results of the unknown loading force test set 1-3, and the second row is the residual fatigue life prediction results of the unknown loading force test set 1-3. The results show that the model has a certain generalization ability.
[0078] Example 2
[0079] Based on Example 1, see Figure 8 This embodiment provides a fatigue crack growth and remaining fatigue life prediction system integrating linear and nonlinear ultrasonic features, which is used to implement the aforementioned fatigue crack growth and remaining fatigue life prediction method integrating linear and nonlinear ultrasonic features. The prediction system includes:
[0080] (1) an ultrasonic monitoring data acquisition module, which is used to obtain ultrasonic guided wave signals of the panel under periodic loading force. The ultrasonic monitoring data acquisition module includes multiple groups of piezoelectric sheets, one part of which is used to excite ultrasonic guided waves, and the other part of which is used to collect ultrasonic guided wave signals;
[0081] (2) a linear and nonlinear feature extraction module, used to calculate the linear ultrasonic feature parameters and the nonlinear ultrasonic feature parameters of each sampling point;
[0082] (3) a feature fusion module, used to calculate the weight corresponding to the sampling point based on the cycle number where the sampling point is located, and weight the linear ultrasonic feature parameter and the nonlinear ultrasonic feature parameter based on the weight;
[0083] (4) A model prediction module is used to predict the crack length and / or residual fatigue life of the plate to be tested based on the weighted linear ultrasonic characteristic parameters and nonlinear ultrasonic characteristic parameters of multiple sampling points using a trained LSTM neural network.
[0084] In summary, the present invention optimizes the combination of linear and nonlinear ultrasonic parameters through a weighted function, and learns features through an LSTM neural network to achieve high-precision prediction of crack propagation length and remaining life. Experimental results show that the present invention has high prediction accuracy and stability over the entire life range, and is suitable for real-time fatigue monitoring and predictive maintenance. By utilizing the linear and nonlinear fusion characteristics of ultrasonic guided wave signals and machine learning, fatigue crack propagation and remaining fatigue life prediction are achieved in plate structures. In addition, the introduction of the first set of data expansion methods increases the amount of data and enables the neural network to make more use of historical data, which solves the contradiction between the amount of historical data and the amount of training data in prediction using the LSTM neural network, and provides a new research idea for solving the problem of using a small amount of data for prediction in engineering applications. The present invention has the following characteristics:
[0085] (1) By utilizing the linear and nonlinear fusion characteristics of ultrasonic guided wave signals and using machine learning, fatigue crack propagation and residual fatigue life prediction can be achieved in plate structures.
[0086] (2) Compared with the traditional single feature prediction, the prediction based on the fusion of linear and nonlinear ultrasonic features can cover the entire life cycle of the material.
[0087] (3) The proposed prediction method has a certain degree of generalization and can adapt to effective prediction under a loading force change of 5%.
[0088] (4) To a certain extent, it solves the contradiction between the amount of training data and historical information, and provides a new research idea for solving the problem of using a small amount of data for prediction in engineering applications.
[0089] Example 3
[0090] This embodiment provides an electronic device, including: one or more processors and a memory, wherein the memory stores one or more programs, and the one or more programs include instructions for executing the fatigue crack propagation and remaining fatigue life prediction method that integrates linear and nonlinear ultrasonic features as described in Example 1.
[0091] like Fig. 9 As mentioned above, at the hardware level, the electronic device includes a processor, an internal bus, a network interface, a memory, and a non-volatile memory, and may also include other hardware required for the business. The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it to achieve the above Figure 1 The fatigue crack growth and remaining fatigue life prediction method integrating linear and nonlinear ultrasonic features. Of course, in addition to the software implementation, the present invention does not exclude other implementations, such as logic devices or a combination of software and hardware, etc., that is, the execution subject of the following processing flow is not limited to each logic unit, but can also be hardware or logic devices.
[0092] The memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.
[0093] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.
[0094] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed by the present invention, and these modifications or replacements should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.
Claims
1. A fatigue crack growth and residual fatigue life prediction method integrating linear and nonlinear ultrasonic features, characterized in that: include: Obtaining the ultrasonic guided wave signal of the plate to be tested under the periodic loading force, and calculating the linear ultrasonic characteristic parameters and nonlinear ultrasonic characteristic parameters of each sampling point; Calculate the weight corresponding to the sampling point based on the cycle number where the sampling point is located, and weight the linear ultrasonic characteristic parameter and the nonlinear ultrasonic characteristic parameter based on the weight; Based on the weighted linear ultrasonic characteristic parameters and nonlinear ultrasonic characteristic parameters of multiple sampling points, the crack length and / or residual fatigue life of the plate to be tested are predicted using the trained LSTM neural network.
2. The fatigue crack growth and residual fatigue life prediction method integrating linear and nonlinear ultrasonic features according to claim 1 is characterized in that: The LSTM neural network learns the relationship between data through a sliding window, and after weighting the linear ultrasonic feature parameters and the nonlinear ultrasonic feature parameters, further includes: Based on the weighted linear ultrasonic characteristic parameters and nonlinear ultrasonic characteristic parameters, the first set of data expansion is achieved by copying the first sampling points of multiple ultrasonic guided wave signals.
3. The fatigue crack growth and residual fatigue life prediction method integrating linear and nonlinear ultrasonic features according to claim 1 is characterized in that: The calculation of the weights corresponding to the sampling points comprises the following steps: The fatigue stage is pre-divided into early, middle and late stages. In the early and middle stages, the weight of the nonlinear ultrasonic characteristic parameters is higher than that of the linear ultrasonic characteristic parameters. In the late stage, the weight of the linear ultrasonic characteristic parameters is higher than that of the early and middle stages.
4. The fatigue crack growth and residual fatigue life prediction method integrating linear and nonlinear ultrasonic features according to claim 3 is characterized in that: The weights corresponding to the sampling points are calculated using a weighting function, where the weighting function is a concave function. In the early stage, the weights of the nonlinear ultrasonic characteristic parameters are greater than 0.9, and in the middle and late stages, the weights of the nonlinear ultrasonic characteristic parameters are greater than 0.
7.
5. The fatigue crack growth and residual fatigue life prediction method integrating linear and nonlinear ultrasonic features according to claim 1 is characterized in that: The weight corresponding to the sampling point is calculated using the following formula: Where N is the number of sampling points, N a is the average total life cycle of all plates, and δ(N) is the weight corresponding to the sampling point.
6. The fatigue crack growth and residual fatigue life prediction method integrating linear and nonlinear ultrasonic features according to claim 1 is characterized in that: The linear ultrasonic characteristic parameter and the nonlinear ultrasonic characteristic parameter are weighted based on the weights using the following formula: β=β×δ(N) A1=A1×(1-δ(N)) Among them, A1 and β are linear ultrasonic characteristic parameters and nonlinear ultrasonic characteristic parameters respectively, and δ(N) is the weight corresponding to the sampling point.
7. The fatigue crack growth and residual fatigue life prediction method integrating linear and nonlinear ultrasonic features according to claim 1 is characterized in that: The invention comprises two LSTM neural networks, namely a first LSTM neural network and a second LSTM neural network, wherein the inputs of the first LSTM neural network and the second LSTM neural network both comprise the number of fatigue cycles, accumulated features, and weighted linear ultrasonic feature parameters and nonlinear ultrasonic feature parameters, the output of the first LSTM neural network is the predicted crack length of the panel to be tested, and the output of the second LSTM neural network is the predicted remaining fatigue life of the panel to be tested.
8. The fatigue crack growth and residual fatigue life prediction method integrating linear and nonlinear ultrasonic features according to claim 1 is characterized in that: The calculation process of the linear ultrasonic characteristic parameters and the nonlinear ultrasonic characteristic parameters of the sampling points includes: Based on the ultrasonic guided wave signal, the fundamental frequency amplitude A1 and the second harmonic amplitude A2 corresponding to the ultrasound are obtained by short-time Fourier transform, and the fundamental frequency amplitude A1 is used as the linear ultrasonic characteristic parameter. as a nonlinear characteristic parameter.
9. The fatigue crack growth and residual fatigue life prediction method integrating linear and nonlinear ultrasonic features according to claim 1 is characterized in that: The LSTM neural network includes multiple LSTM layers and multiple fully connected layers connected in sequence, and for any fully connected layer, a dropout layer is included after the fully connected layer.
10. A fatigue crack growth and residual fatigue life prediction system integrating linear and nonlinear ultrasonic features, characterized in that: A fatigue crack growth and remaining fatigue life prediction method integrating linear and nonlinear ultrasonic features as claimed in any one of claims 1 to 9 is implemented, wherein the prediction system comprises: An ultrasonic monitoring data acquisition module is used to obtain ultrasonic guided wave signals of the panel under periodic loading force, wherein the ultrasonic monitoring data acquisition module includes multiple groups of piezoelectric sheets, a part of which is used to excite ultrasonic guided waves, and another part of which is used to collect ultrasonic guided wave signals; Linear and nonlinear feature extraction module, used to calculate the linear ultrasonic feature parameters and nonlinear ultrasonic feature parameters of each sampling point; A feature fusion module, used to calculate the weight corresponding to the sampling point based on the number of cycles in which the sampling point is located, and weight the linear ultrasonic feature parameter and the nonlinear ultrasonic feature parameter based on the weight; The model prediction module is used to predict the crack length and / or remaining fatigue life of the plate to be tested based on the weighted linear ultrasonic characteristic parameters and nonlinear ultrasonic characteristic parameters of multiple sampling points and using the trained LSTM neural network.