A method and system for predicting early fatigue crack and residual life of a metal structure

Abstract

The application provides a metal structure early fatigue crack and residual life prediction method and system, which comprises the following steps: based on multi-dimensional nonlinear ultrasonic response characteristics, fatigue crack damage measured samples of a typical thickness test piece are established; fatigue damage virtual samples highly consistent with the distribution characteristics of the measured samples are generated to expand the measured sample data; fatigue parameters of the virtual samples are calibrated, and a fatigue crack damage sample library is constructed in combination with the fatigue crack damage measured samples; and based on the fatigue crack damage sample library, an intelligent prediction model is established to quantitatively evaluate the fatigue crack propagation length and the residual service life of the test piece.

Description

Technical Field

[0001] This invention belongs to the field of structural health monitoring technology, and in particular relates to a method and system for predicting early fatigue cracks and remaining life of metal structures. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] Metal structures, due to their excellent strength and machinability, are widely used in aerospace, rail transportation, nuclear power, and wind power. However, under harsh service environments and complex alternating loads, fatigue microcracks are highly susceptible to formation at stress concentration points. These microcracks propagate and eventually form macrocracks, leading to fatigue failure. Studies show that early fatigue performance degradation accounts for 80%-90% of a structure's lifespan, but once macroscopic damage forms, it propagates rapidly, causing sudden structural fracture and failure, and even triggering major safety accidents. Therefore, microcracks have become a significant hidden danger affecting the normal operation of equipment structures.

[0004] Nonlinear ultrasonic guided wave technology can penetrate to the microscopic level and assess the performance degradation of materials based on the nonlinear interaction between waves and damage. This technology essentially reflects the influence of tiny defects on the nonlinearity of materials and is very sensitive to early microstructural (micrometer-level) changes in materials. It has the advantages of being nondestructive and having high detection efficiency, and is an effective means to quantify early fatigue cracks and predict the remaining life of structures.

[0005] Existing nonlinear ultrasonic guided wave fatigue crack prediction techniques mainly include model-based prediction methods and data-driven prediction methods. Model-based prediction methods, through in-depth analysis of structural dynamics and material properties, predict the current operational status of the structure based on physical failure theory. This method typically requires establishing a complex mathematical model of the failure mechanism, comprehensively considering the physical, chemical, and aerodynamic-thermal processes experienced by the structure during service. However, the complexity of its modeling and analysis limits its application and widespread adoption. In contrast, data-driven methods do not rely on structural knowledge. By establishing a sample database and extracting features corresponding to the structural state from sample monitoring data, they can achieve structural fatigue crack damage prediction, making them a current research hotspot in the field of structural health monitoring.

[0006] While data-driven prediction methods offer high application flexibility, they require massive amounts of nonlinear response information as input. However, in real-world fatigue experiments, only one complete nonlinear response data point can be obtained from a single specimen, and the experimental time increases significantly with specimen thickness. Therefore, it is difficult to obtain a large amount of structural nonlinear response data through experiments. Furthermore, the discrepancy between training samples and measured samples is also a significant factor affecting prediction accuracy, but samples collected experimentally are insufficient to cover the thickness range of actual workpieces in use. Summary of the Invention

[0007] To overcome the shortcomings of the prior art, the present invention provides a method for predicting early fatigue cracks and remaining life of metal structures, which can improve the accuracy and reliability of predicting early fatigue crack damage.

[0008] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions:

[0009] In the first aspect, a method for predicting early fatigue cracks and remaining life of metal structures is disclosed, including:

[0010] Based on the multidimensional nonlinear ultrasonic response characteristics, a sample of fatigue crack damage for specimens of typical thickness was established.

[0011] A virtual fatigue damage sample with a distribution characteristic highly consistent with the measured sample is generated to expand the measured damage sample data;

[0012] Fatigue parameters were calibrated on virtual samples, and a fatigue crack damage sample library was constructed by combining them with measured fatigue crack damage samples.

[0013] Based on a fatigue crack damage sample library, an intelligent prediction model is established to quantitatively evaluate the fatigue crack propagation length and remaining service life of the specimen.

[0014] As a further technical solution, based on the multidimensional nonlinear ultrasonic response characteristics, a test sample of fatigue crack damage for specimens of typical thickness is established. The specific steps are as follows:

[0015] During the fracture process of the specimen, the collected ultrasonic guided wave signals were subjected to fast Fourier transform, and the signal amplitudes at the fundamental frequency, second harmonic, third harmonic, and / or difference frequency of the guided wave excitation frequency were extracted.

[0016] The typical multidimensional nonlinear response characteristics of fatigue cracks are calculated based on the signal amplitude extracted from the guided wave excitation frequency.

[0017] Based on typical multidimensional nonlinear response characteristics, fatigue crack length, and remaining life of the specimen, a sample of structural fatigue crack damage was constructed.

[0018] As a further technical solution, the following steps are included before acquiring the ultrasonic guided wave signal of the specimen:

[0019] The specimen was fixed on the fatigue testing machine, and the fatigue loading parameters were set.

[0020] At set intervals, ultrasonic guided wave signals are excited and acquired once, and the detected fatigue crack propagation length is recorded until the specimen fractures.

[0021] As a further technical solution, a virtual fatigue damage sample that highly matches the distribution characteristics of the measured sample is generated, specifically:

[0022] The nonlinear response characteristics of all specimens in the fatigue damage test sample are used as the history curves of the change of fatigue loading cycle as real data.

[0023] Using random noise as input, a generator is used to obtain virtual nonlinear response features that approximate the distribution of measured samples, which are denoted as virtual generated sample features;

[0024] The discriminator is used to distinguish between the input measured sample features and the virtual generated sample features to determine whether the current sample features are "fake data" generated by the generator.

[0025] The generator and discriminator compete with each other, and through continuous iteration and updates, the generator eventually generates virtual generated samples that are highly consistent with the distribution of the measured samples.

[0026] As a further technical solution, fatigue parameters are calibrated on the virtual samples, and a fatigue crack damage sample library is constructed by combining the virtual samples with measured fatigue crack damage samples. Specifically:

[0027] The multidimensional nonlinear response characteristic curve of the measured sample is used as the input of the MLSTM network, and the "sample thickness, crack length and remaining life" corresponding to the nonlinear characteristics are used as the model output. The "nonlinear response characteristic-fatigue parameter" mapping model is established using the MLSTM network.

[0028] The nonlinear response characteristic curves in the generated samples are input into the MLSTM model to obtain the fatigue parameters corresponding to the nonlinear characteristics of the generated samples, including: specimen thickness, crack length, and fatigue life.

[0029] A fatigue crack damage sample library was constructed based on measured and generated samples. The specimen thickness and nonlinear response characteristics are the data features of the sample library, while the fatigue crack propagation length and specimen remaining life are the data labels of the sample library.

[0030] As a further technical solution, an intelligent prediction model is established based on a fatigue crack damage sample library, specifically including:

[0031] The fatigue crack damage sample library is used as the training sample. The data features in the fatigue crack damage sample library are used as input and the data labels are used as output. A mapping model between "specimen thickness and nonlinear characteristics" and "crack propagation length and specimen remaining life" is established using a BPNN network.

[0032] As a further technical solution, the data characteristics in the fatigue crack damage sample library include specimen thickness and nonlinear response characteristics;

[0033] The data labels include fatigue crack propagation length and specimen remaining life.

[0034] As a further technical solution, a prediction step is also included: when predicting the fatigue damage state of a structure, the thickness of the test piece and the typical nonlinear response characteristics of the current state calculated based on the collected ultrasonic guided waves are input into the trained BPNN prediction model, so as to obtain the fatigue crack propagation length and remaining service life of the structure under the current state.

[0035] Secondly, a system for predicting early fatigue cracks and remaining life of metal structures is disclosed, including:

[0036] The fatigue crack damage test sample establishment module is configured to: establish fatigue crack damage test samples of specimens with typical thickness based on multidimensional nonlinear ultrasonic response characteristics.

[0037] The damage measurement sample data augmentation module is configured to: generate a fatigue damage virtual sample that is highly consistent with the distribution characteristics of the measured sample, and augment the damage measurement sample data.

[0038] The fatigue crack damage sample library establishment module is configured to: calibrate the fatigue parameters of virtual samples and combine them with measured fatigue crack damage samples to construct a fatigue crack damage sample library;

[0039] The intelligent prediction model building module is configured to: build an intelligent prediction model based on the fatigue crack damage sample library, which is used to quantitatively evaluate the fatigue crack propagation length and remaining service life of the specimen.

[0040] The above one or more technical solutions have the following beneficial effects:

[0041] The technical solution of this invention is a data-driven prediction method based on artificial intelligence networks, which can improve the accuracy and reliability of predicting early fatigue crack damage in structures.

[0042] The technical solution of this invention utilizes GAN and MLSTM networks to quickly and cost-effectively obtain any number of virtual generated samples that are highly consistent with the feature distribution of measured fatigue damage sample data, thereby establishing a rich fatigue crack damage sample library to meet the needs of data-driven prediction methods for massive data. By utilizing multidimensional nonlinear response characteristics under different fatigue loading cycles, structural thickness, fatigue crack propagation length, remaining service life, and other parameters, a fatigue damage prediction model is established based on a BPNN network. This can reduce the dependence on the continuity of input features in the time series and improve the accuracy and efficiency of real-time prediction of fatigue crack damage state.

[0043] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0044] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0045] Figure 1 This is an embodiment of the metal structure fatigue crack detection system of the present invention;

[0046] Figure 2 This is a schematic diagram illustrating the quantitative assessment of structural fatigue crack propagation length and prediction of remaining life in an embodiment of the present invention. Detailed Implementation

[0047] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0048] It should be noted that the terminology used herein is for the purpose of describing particular implementations only and is not intended to limit the exemplary implementations of the present invention.

[0049] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0050] The overall concept proposed in this invention is as follows:

[0051] This disclosure utilizes artificial intelligence networks to establish a data-driven model for quantitative assessment of early fatigue crack propagation length and prediction of remaining service life in structures. First, fatigue loading experiments are conducted to obtain multidimensional nonlinear response characteristics, including high-order harmonics and mixing, from a small number of specimens of typical thickness. These characteristics are then combined with the crack propagation length and remaining service life variation history to construct a fatigue damage test sample. Next, a generative sample highly consistent with the feature distribution of the test sample is obtained using a generative adversarial network (GAN) and a multi-layer long short-term memory (MLSTM) network. This expands the fatigue damage data to include multidimensional nonlinear response characteristics, establishing a fatigue damage sample database. Finally, a backpropagation neural network (BPNN) is combined with the fatigue damage samples to establish a model for quantitative assessment of fatigue crack propagation length and prediction of remaining service life based on specimen thickness and multidimensional nonlinear response characteristics. In practical applications, simply inputting the thickness of the test specimen and the measured damage characteristics into the BPNN prediction model yields the fatigue crack propagation length and remaining service life of the structure in its current state.

[0052] Example 1

[0053] This embodiment discloses a fatigue crack detection system for metal structures; see attached document. Figure 1 As shown, the main components include aluminum alloy specimens, piezoelectric transducers (PZT), universal fatigue testing machines, digital microscopes, nonlinear ultrasonic guided wave testing instruments, and host computers.

[0054] (1) 6061-T6 aluminum alloy specimens with typical thicknesses (2mm, 4mm, 8mm) were selected as the research object. Before the fatigue test began, an initial crack with a length of 5mm was artificially pre-fabricated in the middle of the specimen.

[0055] (2) Three PZT sensors were attached to the surface of each specimen, two of which were used as high-frequency sensors (frequency ω). a ), low frequency (frequency is w) b A sinusoidal modulated waveguide exciter, and a waveguide receiver;

[0056] (3) The universal fatigue testing machine is responsible for applying cyclic loads to aluminum alloy specimens;

[0057] (4) A digital microscope is used to observe the length of fatigue crack propagation in real time;

[0058] (5) The nonlinear ultrasonic guided wave testing instrument is connected to the PZT sensor and is responsible for the transmission of guided wave signals;

[0059] (6) The host computer is responsible for generating the excitation signal and displaying, storing and processing the received signal.

[0060] Example 2

[0061] See appendix Figure 2 As shown, based on the aforementioned metal structure fatigue crack detection system, this embodiment discloses a method for predicting early fatigue cracks and remaining life of metal structures, including:

[0062] Step 1: Based on the multidimensional nonlinear ultrasonic response characteristics, establish a sample of fatigue crack damage for aluminum alloy specimens of typical thickness:

[0063] (1) Fix the specimen on the universal fatigue testing machine and set the fatigue loading parameters such as fatigue loading cycle (Hz), waveform, stress ratio, and maximum stress;

[0064] (2) Every 2×10 4 One cycle is performed to excite and acquire the ultrasonic guided wave signal, and the fatigue crack propagation length detected by the digital microscope is recorded until the specimen breaks.

[0065] (3) Perform a Fast Fourier Transform (FFT) on the acquired ultrasonic guided wave signal and extract the fundamental frequency (w) of the guided wave excitation frequency. a With w b ), double frequency (2W) a With 2w b ), triple frequency (3W) a With 3w b ), sum frequency, difference frequency (w) a+b With w a-b The signal amplitudes under these conditions are denoted as follows: and

[0066] Here, the sum frequency is the sum of the two fundamental frequencies, i.e., w. a+b The difference frequency is the difference between two fundamental frequencies, i.e., w. a-b .

[0067] (4) Calculate the typical multidimensional nonlinear response characteristics of fatigue cracks: and

[0068] Existing research has shown that these six nonlinear parameters change significantly with varying crack length, so they are used to characterize crack length. The advantage is that these nonlinear response parameters are strongly correlated with crack propagation length, resulting in high prediction accuracy for models built upon them.

[0069] (5) Multidimensional nonlinear ultrasonic response characteristics calculated based on typical thickness specimens under different stress ratios The measured fatigue crack propagation length, remaining life of the specimen, and typical thickness of the specimen were used to construct a sample of the actual fatigue crack damage.

[0070] The remaining life of the specimen is calculated by subtracting the current loading period from the fatigue loading period the specimen can withstand. The fatigue loading period the specimen can withstand is the total loading period up to the moment the specimen fractures, which can only be determined at the time of fracture.

[0071] The data in the sample mainly consists of two parts: damage features, composed of specimen thickness and nonlinear ultrasonic response characteristics (including six parameters); and damage labels, composed of crack length and remaining life. The damage features are the model input, and the damage labels are the model output. There is a one-to-one relationship between the input and output. Each row in the sample table contains thickness, features, length, and life, with different rows representing different stages.

[0072] Step 2: Leveraging the powerful feature learning and generation capabilities of GAN networks, through continuous adversarial and iterative interaction between the generator and the recognizer, virtual fatigue damage samples are generated that highly match the multidimensional nonlinear ultrasonic response characteristics of the measured damage samples, thereby expanding the damage sample data.

[0073] (1) Nonlinear response characteristics of all specimens in the fatigue damage test samples and The history curve showing the change with the fatigue loading cycle is used as the real data, i.e. Figure 2 The actual test samples were used to train the GAN discriminator, the MLSTM model, and the BPNN model.

[0074] (2) Using random noise as the input generator, the generator of the GAN network is used to obtain virtual nonlinear response characteristics that approximate the distribution of the measured samples. and Let these be denoted as virtual generated sample features;

[0075] (3) Use the discriminator of the GAN network to discriminate the nonlinear response characteristics of all specimens in the input measured samples and the generated sample characteristics to determine whether the current sample characteristics are "fake data" generated by the generator.

[0076] (4) The generator and discriminator compete against each other. Through continuous iteration and updates, the generator of the GAN network can eventually generate virtual generated sample features that are highly consistent with the feature distribution of the measured samples. Each sample data contains and Typical nonlinear response history curves.

[0077] Step 3: Based on the high accuracy of MLSTM network in identifying time series data features, fatigue parameters are calibrated for the features of the virtually generated samples. Then, combined with measured fatigue crack damage samples, a rich fatigue crack damage sample library is constructed.

[0078] (1) The multidimensional nonlinear response characteristics of the measured samples and The curve is used as the input to the MLSTM network, and the "specimen thickness, crack length, and remaining life" corresponding to the nonlinear characteristics are used as the model output. The MLSTM network is used to establish a "nonlinear response characteristics-fatigue parameters" mapping model.

[0079] The mapping model has a fixed structure. Only the parameters need to be initially set. Then, the nonlinear response characteristics in the measured sample are given to the input of the model. The output is "the specimen thickness, crack length, and remaining life". Crack length and remaining life can be collectively referred to as fatigue parameters. The parameters and structure of the model correspond to a model.

[0080] (2) Nonlinear response features in the generated samples and The curve is input into the MLSTM model, i.e. the mapping model, to obtain fatigue parameters such as specimen thickness, crack length, and fatigue life corresponding to the nonlinear characteristics of the generated sample.

[0081] During the fatigue loading process, the specimen has a set of nonlinear response characteristics (containing six values) at each moment. Therefore, there are N sets of nonlinear response characteristics (containing n×6 values) during the entire loading process. By converting the N values ​​of the same parameter into a curve according to the time history, the six parameters can be used to obtain six curves, namely the nonlinear response characteristic curves.

[0082] (3) A fatigue crack damage sample library was constructed based on measured samples and generated samples, including specimen thickness and nonlinear response characteristics. and The fatigue crack propagation length and the remaining life of the specimen are the data characteristics of the sample library.

[0083] Among them, the specimen thickness and nonlinear response characteristics are attached. Figure 2 The inputs to the BPNN model are fatigue crack propagation length and specimen remaining life, while the outputs are fatigue crack propagation length and specimen remaining life. Ultimately, the entire patent completes the function of predicting fatigue crack propagation length and specimen remaining life based on two known parameters: specimen thickness and nonlinear response characteristics.

[0084] Step 4: Utilizing the excellent self-learning and generalization capabilities of BPNN networks, establish an intelligent prediction model for the quantitative assessment of fatigue crack propagation length and remaining service life based on specimen thickness and multidimensional nonlinear ultrasonic response characteristics.

[0085] (1) The fatigue crack damage sample library was used as the training sample, and the data characteristics in the sample (specimen thickness, nonlinear response characteristics) were used. and Using data labels (fatigue crack propagation length and specimen remaining life) as input and data labels (fatigue crack propagation length and specimen remaining life) as output, a mapping model between "specimen thickness and nonlinear characteristics" and "crack propagation length and specimen remaining life" is established using a BPNN network.

[0086] The above mapping model consists of an input layer, a hidden layer, and an output layer. The input layer contains seven neurons, corresponding to the specimen thickness and six nonlinear response features, respectively. The hidden layer contains 100 neurons. The output layer contains two neurons, corresponding to the crack propagation length and the remaining lifetime of the specimen, respectively. The model loss function is E = (√2 * (|predicted crack length / actual crack length|)) * (√2 * (predicted crack length / actual crack length))) 2 +|Predicted Remaining Life - Actual Remaining Life|2 The goal of model training is to obtain model parameters that satisfy the condition E < 0.001, including connection weights and thresholds between the input layer and hidden layers, and connection weights and thresholds between the hidden layer and the output layer.

[0087] (2) When predicting the fatigue damage state of a structure, the thickness of the test piece and the typical nonlinear response characteristics of the current state calculated based on the collected ultrasonic guided waves are input into the trained BPNN prediction model, i.e., the mapping model, so that the length of the fatigue crack and the remaining service life of the structure under the current state can be obtained.

[0088] The present invention addresses the problem of quantitative assessment of fatigue crack propagation length and prediction of remaining life by proposing an ultrasonic detection method based on multidimensional nonlinear characteristics. By fusing the high-order harmonics and mixed-frequency response of guided wave signals, the reliability and accuracy of crack damage detection are improved.

[0089] The technical solution of this invention utilizes the powerful feature and learning capabilities of GAN networks to obtain virtual generated samples with a high degree of consistency in feature distribution of any number of fatigue damage test samples with typical thickness specimens at low cost and fast speed, thereby expanding fatigue crack damage sample data under different conditions.

[0090] The technical solution of this invention utilizes the powerful interpretation capability of MLSTM network for time series features, and establishes a "nonlinear feature-fatigue parameter" mapping model by combining measured fatigue damage samples. This model can accurately calibrate the fatigue parameters corresponding to the nonlinear response feature curves in the generated samples, and then establish a rich fatigue crack damage sample library by combining the measured sample features and fatigue parameters.

[0091] The technical solution of this invention utilizes abundant fatigue crack damage sample data and establishes a crack length and life prediction model based on a BPNN network. This can reduce the dependence of the prediction results on the feature time series, and can achieve crack propagation length assessment and remaining life prediction based solely on the nonlinear damage characteristics of a specimen of a specific thickness at the current moment.

[0092] Example 2

[0093] The purpose of this embodiment is to provide a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the above-described method.

[0094] Example 3

[0095] The purpose of this embodiment is to provide a computer-readable storage medium.

[0096] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the steps of the above method.

[0097] Example 4

[0098] The purpose of this embodiment is to provide a system for predicting early fatigue cracks and remaining life of metal structures, including:

[0099] The fatigue crack damage test sample establishment module is configured to: establish fatigue crack damage test samples of specimens with typical thickness based on multidimensional nonlinear ultrasonic response characteristics.

[0100] The damage measurement sample data augmentation module is configured to: generate a fatigue damage virtual sample that is highly consistent with the distribution characteristics of the measured sample, and augment the damage measurement sample data.

[0101] The fatigue crack damage sample library establishment module is configured to: calibrate the fatigue parameters of virtual samples and combine them with measured fatigue crack damage samples to construct a fatigue crack damage sample library;

[0102] The intelligent prediction model building module is configured to: build an intelligent prediction model based on the fatigue crack damage sample library, which is used to quantitatively evaluate the fatigue crack propagation length and remaining service life of the specimen.

[0103] The steps and methods involved in the apparatuses of Embodiments 2, 3, and 4 above correspond to those in Embodiment 1. For specific implementation details, please refer to the relevant description section of Embodiment 1. The term "computer-readable storage medium" should be understood as a single medium or multiple media including one or more instruction sets; it should also be understood as including any medium capable of storing, encoding, or carrying an instruction set for execution by a processor and enabling the processor to perform any of the methods in this invention.

[0104] Those skilled in the art will understand that the modules or steps of the present invention described above can be implemented using general-purpose computer devices. Optionally, they can be implemented using computer-executable program code, thereby allowing them to be stored in a storage device for execution by a computer device, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. The present invention is not limited to any particular combination of hardware and software.

[0105] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A method for predicting early fatigue cracks and remaining life of metal structures, characterized in that, include: Based on the multidimensional nonlinear ultrasonic response characteristics, a sample of fatigue crack damage for specimens of typical thickness was established. A virtual fatigue damage sample with a distribution characteristic highly consistent with the measured sample is generated to expand the measured damage sample data; Fatigue parameters are calibrated on virtual samples, and a fatigue crack damage sample library is constructed by combining virtual samples with measured fatigue crack damage samples. Specifically: The multidimensional nonlinear response characteristic curve of the measured sample is used as the input of the MLSTM network, and the specimen thickness, crack length and remaining life corresponding to the nonlinear characteristics are used as the model output. The nonlinear response characteristic-fatigue parameter mapping model is established using the MLSTM network. The nonlinear response characteristic curves in the generated samples are input into the MLSTM model to obtain the fatigue parameters corresponding to the nonlinear characteristics of the generated samples, including: specimen thickness, crack length, and fatigue life. A fatigue crack damage sample library was constructed based on measured samples and generated samples. The specimen thickness and nonlinear response characteristics are the data features of the sample library, and the fatigue crack propagation length and specimen remaining life are the data labels of the sample library. Based on a fatigue crack damage sample library, an intelligent prediction model is established to quantitatively evaluate the fatigue crack propagation length and remaining service life of specimens. Specifically, the intelligent prediction model based on the fatigue crack damage sample library includes: The fatigue crack damage sample library was used as the training sample. The data features in the fatigue crack damage sample library were used as the input and the data labels were used as the output. A mapping model between specimen thickness, nonlinear features and crack propagation length and specimen remaining life was established using a BPNN network.

2. The method for predicting early fatigue cracks and remaining life of a metal structure as described in claim 1, characterized in that, Based on the multidimensional nonlinear ultrasonic response characteristics, a sample of fatigue crack damage for specimens of typical thickness was established. The specific steps are as follows: During the fracture process of the specimen, the collected ultrasonic guided wave signals were subjected to fast Fourier transform, and the signal amplitudes at the fundamental frequency, second harmonic, third harmonic, and / or difference frequency of the guided wave excitation frequency were extracted. The typical multidimensional nonlinear response characteristics of fatigue cracks are calculated based on the signal amplitude extracted from the guided wave excitation frequency. Based on typical multidimensional nonlinear response characteristics, fatigue crack length, and remaining life of the specimen, a sample of structural fatigue crack damage was constructed.

3. The method for predicting early fatigue cracks and remaining life of a metal structure as described in claim 1, characterized in that, Before acquiring the ultrasonic guided wave signal of the specimen, the following steps are also included: The specimen was fixed on the fatigue testing machine, and the fatigue loading parameters were set. At set intervals, ultrasonic guided wave signals are excited and acquired once, and the detected fatigue crack propagation length is recorded until the specimen fractures.

4. The method for predicting early fatigue cracks and remaining life of a metal structure as described in claim 1, characterized in that, Generating virtual fatigue damage samples that highly match the distribution characteristics of measured samples is specifically as follows: The nonlinear response characteristics of all specimens in the fatigue damage test sample are used as the history curves of the change of fatigue loading cycle as real data. Using random noise as input, a generator is used to obtain virtual nonlinear response features that approximate the distribution of measured samples, which are denoted as virtual generated sample features; The discriminator is used to distinguish between the input real sample features and the virtual generated sample features to determine whether the current sample features are fake data generated by the generator. The generator and discriminator compete with each other, and through continuous iteration and updates, the generator eventually generates virtual generated samples that are highly consistent with the distribution of the measured samples.

5. The method for predicting early fatigue cracks and remaining life of a metal structure as described in claim 1, characterized in that, It also includes a prediction step: when predicting the fatigue damage state of a structure, the thickness of the test piece and the typical nonlinear response characteristics of the current state calculated based on the collected ultrasonic guided waves are input into the trained BPNN prediction model, so as to obtain the fatigue crack propagation length and remaining service life of the structure under the current state.

6. A system for predicting early fatigue cracks and remaining life of metal structures, characterized in that, include: The fatigue crack damage test sample establishment module is configured to: establish fatigue crack damage test samples of specimens with typical thickness based on multidimensional nonlinear ultrasonic response characteristics. The damage measurement sample data augmentation module is configured to: generate a fatigue damage virtual sample that is highly consistent with the distribution characteristics of the measured sample, and augment the damage measurement sample data. The fatigue crack damage sample library creation module is configured to: calibrate fatigue parameters on virtual samples and, in conjunction with measured fatigue crack damage samples, construct a fatigue crack damage sample library, specifically: The multidimensional nonlinear response characteristic curve of the measured sample is used as the input of the MLSTM network, and the specimen thickness, crack length and remaining life corresponding to the nonlinear characteristics are used as the model output. The nonlinear response characteristic-fatigue parameter mapping model is established using the MLSTM network. The nonlinear response characteristic curves in the generated samples are input into the MLSTM model to obtain the fatigue parameters corresponding to the nonlinear characteristics of the generated samples, including: specimen thickness, crack length, and fatigue life. A fatigue crack damage sample library was constructed based on measured samples and generated samples. The specimen thickness and nonlinear response characteristics are the data features of the sample library, and the fatigue crack propagation length and specimen remaining life are the data labels of the sample library. The intelligent prediction model building module is configured to: build an intelligent prediction model based on a fatigue crack damage sample library to quantitatively evaluate the fatigue crack propagation length and remaining service life of the specimen; the intelligent prediction model based on the fatigue crack damage sample library specifically includes: The fatigue crack damage sample library was used as the training sample. The data features in the fatigue crack damage sample library were used as the input and the data labels were used as the output. A mapping model between specimen thickness, nonlinear features and crack propagation length and specimen remaining life was established using a BPNN network.

7. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the method described in any one of claims 1-5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it performs the steps of the method described in any one of claims 1-5 above.

Images