Nonlinear ultrasonic based micro-crack damage assessment and life prediction method for welded structures

Abstract

The present application relates to the technical field of nondestructive testing of welded structures, and provides a welded structure micro-crack damage evaluation and life prediction method based on nonlinear ultrasonic waves, comprising: calculating the dispersion curve of ultrasonic Lamb waves in the welded structure; selecting appropriate Lamb wave modes based on the dispersion curve, preparing welded structure test pieces with different micro-crack damage degrees through fatigue loading, exciting Lamb waves through a determined wedge incidence angle by using an ultrasonic guided wave detection system, enhancing ultrasonic signals by using pulse inversion and empirical mode decomposition; and performing short-time Fourier transform on the enhanced signals to extract the second harmonic component in the signals and calculate nonlinear ultrasonic response parameters; constructing a "damage state-nonlinear ultrasonic parameter-residual life" data set, outputting the fatigue residual life of the welded structure with an unknown damage state through the trained model, and realizing quantitative evaluation of the micro-crack damage degree of the welded structure and residual life prediction.

Description

Technical Field

[0001] This invention relates to the technical field of nondestructive testing of welded structures, and in particular to a method for assessing microcrack damage and predicting lifespan of welded structures based on nonlinear ultrasound. Background Technology

[0002] Welded structures, as key connection and load-bearing units in major equipment and engineering systems, are widely used in fields such as nuclear power, aerospace, energy, and high-end manufacturing. Examples include pressure-bearing pipelines and pressure boundary welds in nuclear power equipment, and aluminum alloy panels and module connection welds in aerospace equipment. These welded structures typically operate under conditions of high temperature, high pressure, alternating loads, vibration and shock, or complex environments. Their structural integrity and service life are directly related to the safe operation and reliability of the equipment system.

[0003] During long-term service, welded structures inevitably experience fatigue loads, leading to early damage phenomena such as microcracks, interfacial micro-damage, and localized material degradation gradually forming in the weld and its heat-affected zone. Studies have shown that the early microcrack initiation and stable propagation stage of materials and structures typically accounts for approximately 70% to 80% of the total fatigue life, making it a critical stage determining the remaining life and failure risk of the structure. However, the damage scale at this stage is often at the micrometer or even submicrometer level, without forming obvious geometric discontinuities, posing a significant challenge to nondestructive testing.

[0004] Currently, linear ultrasonic testing remains a crucial method for non-destructive testing of welded structures in engineering applications. It primarily relies on the linear reflection characteristics of ultrasonic waves at material interfaces or defects for defect identification. However, when the damage in a welded structure is mainly characterized by early microcracks or microscale contact damage, their size is much smaller than the ultrasonic testing wavelength, resulting in extremely weak linear reflection signals. These signals are easily drowned out by scattering noise from the material's internal structure or background echoes, leading to missed detections or misjudgments. Furthermore, welded joint areas typically exhibit non-uniform microstructure and significant gradients in mechanical properties, further reducing the sensitivity of linear ultrasonic testing for early micro-damage.

[0005] In contrast, nonlinear ultrasonic testing technology, based on the nonlinear elastic response mechanism caused by material damage, exhibits significantly higher sensitivity to early-stage damage such as microcracks. Microcracks, under alternating ultrasonic stress, exhibit periodic contact and separation behavior, the so-called "microcrack breathing effect," which leads to significant harmonic components and nonlinear modulation phenomena during ultrasonic wave propagation. By extracting characteristic parameters such as second harmonics and nonlinear ultrasonic parameters, the microcrack damage state within the material can be effectively characterized before macroscopic cracks form. Existing research has shown a good correlation between nonlinear ultrasonic response and material fatigue damage evolution, making it particularly suitable for the detection and assessment of early and mid-stage micro-damage in welded structures.

[0006] Although nonlinear ultrasound has significant advantages in microcrack detection, existing research mainly focuses on qualitative damage identification or single-state discrimination, making it difficult to quantitatively assess the degree of microcrack damage in welded structures. Furthermore, a systematic approach combining nonlinear ultrasound detection results with structural fatigue life prediction is lacking. In high-reliability equipment fields such as nuclear power and aerospace, welded structures often have long service lives and high safety margin requirements, urgently necessitating a detection technology capable of predicting and assessing the remaining life of welded structures based on early micro-damage information.

[0007] Therefore, how to fully utilize the high sensitivity of nonlinear ultrasound to microcrack damage, establish a quantitative relationship between microcrack damage evolution and nonlinear ultrasonic response, and further realize reliable prediction of the fatigue remaining life of welded structures, under the premise of being applicable to welded structures in different engineering fields, has become an important technical problem that urgently needs to be solved in the field of nondestructive testing and structural health monitoring of welded structures. Summary of the Invention

[0008] To overcome the problems of existing technologies in nondestructive testing of welded structures, such as insensitivity to early microcrack damage, difficulty in reliably identifying microcracks before macrocrack formation, and the limited applicability of existing test results to predict the remaining life of welded structures, this invention proposes a method for assessing microcrack damage and predicting the life of welded structures based on nonlinear ultrasound. This method fully utilizes the high sensitivity of nonlinear ultrasound to the microcrack breathing effect. By extracting nonlinear ultrasonic response parameters of the welded structure during fatigue damage evolution, it achieves quantitative characterization of early and mid-stage microcrack damage and further constructs a mapping relationship between damage state and remaining fatigue life, thereby enabling reliable prediction of the welded structure's life.

[0009] This invention focuses on the early identification and life prediction of microcrack damage in welded structures. Addressing the limitations of traditional linear ultrasonic testing in effectively characterizing microscale damage and providing only qualitative assessments, it introduces a detection and analysis method based on nonlinear ultrasonic guided waves. By analyzing the harmonic components generated during ultrasonic wave propagation in the welded structure, and utilizing the nonlinear acoustic response of microcracks under alternating stress, sensitive detection of microcrack damage within the welded structure is achieved. Simultaneously, by combining fatigue loading test data, an intrinsic relationship is established between the nonlinear ultrasonic response and the damage evolution and remaining life of the welded structure. This allows the detection results to not only reflect the current damage state but also be used for life assessment and prediction.

[0010] To achieve the above objectives, this invention provides a method for assessing microcrack damage and predicting the lifespan of welded structures based on nonlinear ultrasound, the method comprising: S1. Based on the material type, structural thickness and known acoustic properties of the welded structure, calculate the dispersion curve of ultrasonic Lamb wave in the welded structure, and obtain the distribution law of phase velocity and group velocity of different order Lamb wave modes with frequency. S2, based on the dispersion curve, select a suitable Lamb wave mode pair within the preset detection frequency range, and determine the center frequency, phase velocity, and wedge incident angle required to excite the mode corresponding to the mode pair; S3, welded structural specimens with different microcrack damage levels were prepared by fatigue loading. Lamb waves were excited by an ultrasonic guided wave detection system through a determined wedge incident angle. The pulse inversion method was used to detect the specimens with different damage states and collect the pulse inversion synthesized signal. S4. The pulse inversion synthesized signal is enhanced with ultrasound signal by empirical mode decomposition; and the enhanced pulse inversion synthesized signal is subjected to short-time Fourier transform to obtain time-frequency diagram, and the amplitude of the corresponding frequency band is extracted to characterize the fundamental frequency component and the second harmonic component, and the nonlinear ultrasound response parameters are calculated based on the fundamental frequency component and the second harmonic component. S5. The fatigue loading cycles and nonlinear ultrasonic response parameters corresponding to welded structural specimens with different crack damage states are used to construct a dataset of "damage state-nonlinear ultrasonic parameters-remaining life", and the dataset is normalized. S6. Construct an artificial neural network model, using fatigue cycles and normalized nonlinear ultrasonic response parameters as model inputs, and the remaining fatigue life of the welded structure as model output. Train and optimize the artificial neural network model using the normalized dataset. S7 performs nonlinear ultrasonic Lamb wave detection on welded structures with unknown damage states, extracts their nonlinear ultrasonic response parameters, inputs the fatigue cycles and normalized nonlinear ultrasonic response parameters into a trained artificial neural network model, and outputs the fatigue remaining life from the artificial neural network model, thereby realizing the quantitative assessment of the degree of microcrack damage and the prediction of the remaining life of the welded structure.

[0011] Furthermore, the specific steps for calculating the dispersion curve of the ultrasonic Lamb wave in the welded structure in step S1 are as follows: Assume the Lamb wave propagates horizontally along the plate surface, and the phase velocity of the Lamb wave is... The following text is abbreviated as Frequency-thickness product range (MHz·mm); Define the normalized wavenumber component of the transverse wave in the thickness direction (perpendicular to the propagation direction). for:

[0012] Define the normalized wavenumber component of the longitudinal wave in the thickness direction (perpendicular to the propagation direction). for:

[0013] In the formula, The transverse wave velocity; For longitudinal wave velocity; Characteristic equations of the symmetric mode (S-mode):

[0014] Characteristic equation of antisymmetric mode (A-mode):

[0015] For each frequency-thickness product In the phase velocity range Internal scan, searching and The roots are then determined using numerical methods such as the bisection method to obtain the precise roots, yielding the phase velocities of each order of symmetric and antisymmetric modes. ; Calculating group velocity from phase velocity using dispersion relation :

[0016] Finally, based on the known plate thickness The curves of phase velocity / group velocity as a function of frequency were obtained.

[0017] Furthermore, in step S2, the wedge incident angle The formula for calculation is:

[0018] In the formula, This indicates the longitudinal wave velocity within the wedge.

[0019] Preferably, in step S3, fatigue loading adopts a constant amplitude cyclic loading method, and the damage state of the welded structure corresponding to each loading cycle is recorded.

[0020] Furthermore, the specific steps in step S3 for detecting specimens in different damage states and acquiring the original ultrasonic guided wave signals using the pulse inversion method are as follows: For the same detection location, two sets of Lamb wave excitation signals with a phase difference of 180° were applied sequentially, and the corresponding time-domain response signals were acquired respectively. and :

[0021] in, For the amplitude of the excitation signal, The fundamental angular frequency, For time; The two sets of time-domain response signals are superimposed to obtain the pulse inversion composite signal. :

[0022] According to the nonlinear ultrasound propagation theory, pulse inversion synthesized signal The fundamental frequency and odd harmonics are effectively suppressed, while the second harmonic component is enhanced.

[0023] Furthermore, the specific steps for enhancing the acquired original ultrasonic guided wave signal through empirical mode decomposition in step S4 are as follows: Due to the non-uniform microstructure and strong scattering attenuation caused by the coarse-grained structure in the weld region, the nonlinear second harmonic component in the directly acquired pulse inversion synthesized signal is extremely weak and difficult to extract effectively. Therefore, this invention first performs empirical mode decomposition enhancement processing on the acquired pulse inversion synthesized signal before the short-time Fourier transform (STFT) to effectively improve the second harmonic signal-to-noise ratio.

[0024] Based on the time-scale characteristics of the signal itself, the pulse is inverted to synthesize the signal. Adaptively decomposed into several intrinsic mode function (IMF) components With a residual component :

[0025] Specifically, the process involves first identifying the pulse inversion synthesized signal. Construct the upper envelope for each of the local maxima and minima. and lower envelope Calculate the local mean ; Then synthesize the signal from the pulse inversion. Subtract local mean from middle get ; Repeat the above process until Satisfying intrinsic mode function (IMF) components The two conditions that need to be met are: obtaining the first intrinsic mode function (IMF) component. ;use As a new signal to be decomposed, continue decomposition until the residual components are obtained. It is a monotonic function or a constant; The intrinsic mode function (IMF) components obtained from the decomposition Calculate its relationship with the theoretical second harmonic reference signal. Normalized correlation coefficient between :

[0026] Selecting the normalized correlation coefficient Largest intrinsic mode function (IMF) component As an effective component containing the second harmonic principal energy, the intrinsic mode function (IMF) component... The enhanced second harmonic signal is obtained by reconstructing the low-frequency trend term in the residual component. :

[0027] Furthermore, each intrinsic mode function (IMF) component The following two conditions must be met: (1) Over the entire signal length, the number of extreme points and the number of zero crossings are equal or at most differ by one; (2) At any time, the mean of the upper envelope defined by the local maxima and the lower envelope defined by the local minima is zero.

[0028] Furthermore, in step S4, the enhanced second harmonic signal... Perform a short-time Fourier transform to obtain a time-frequency graph, and extract the amplitude of the corresponding frequency band to characterize the fundamental frequency component. Second harmonic component And based on the fundamental frequency component With second harmonic components Calculation of nonlinear ultrasonic response parameters : .

[0029] Furthermore, in step S6, the artificial neural network model is a feedforward neural network, a convolutional neural network, or a combination thereof, including an input layer, a first LSTM layer, a second LSTM layer, a first connection layer, a second connection layer, and an output layer connected in sequence, with each layer connected one after another, and data being passed layer by layer in a unidirectional order. The input layer includes a multi-channel temporal input unit group, with each feature corresponding to an independent temporal input channel. The fatigue cycles and normalized nonlinear ultrasonic response parameters are recombined by the multi-channel temporal input unit to generate a temporal tensor. The first LSTM layer consists of multiple stacked LSTM memory units. Each memory unit integrates four core gating structures: a forget gate, an input gate, an output gate, and a cell state. The gating units uniformly adopt the Sigmoid activation function, and the cell state update is combined with the Tanh activation function to complete the nonlinear transformation. The temporal tensor is extracted by the gating structure to obtain a shallow temporal feature tensor, which is used to capture the long-distance temporal dependencies of the entire fatigue loading process, extract the basic temporal change law of the micro-damage evolution of the component, filter and monitor random interference, and retain the basic temporal information of fatigue damage evolution with each cycle. The second LSTM layer consists of multiple stacked LSTM memory units. Each memory unit integrates four core gating structures: a forget gate, an input gate, an output gate, and a cell state. The gating units uniformly adopt the Sigmoid activation function, and the cell state update is combined with the Tanh activation function to complete the nonlinear transformation. The shallow temporal feature tensor undergoes multi-layer gating iterative operations to complete deep feature abstraction based on the shallow temporal information, explore the high-order coupling correlation between fatigue cycles and nonlinear ultrasound multi-parameters, filter redundant temporal information, and output the aggregated high-dimensional deep feature tensor. The first connection layer is a standard fully connected network structure, consisting of neuron weight matrices, bias terms, and ReLU nonlinear activation functions. The high-dimensional feature tensor first undergoes linear weighting operations, and then a nonlinear transformation is achieved through the activation function. Simultaneously, the high-dimensional features are compressed and redundant features are removed, outputting a low-dimensional fused feature vector. This initially establishes a nonlinear mapping relationship between temporal damage features and welding fatigue remaining life. The second connection layer is a fully connected network structure. The low-dimensional fused feature vectors are subjected to linear weighting and nonlinear fitting again to further complete the feature dimension reduction, correct the previous mapping bias, refine and optimize the regression mapping curve from damage features to fatigue life, and generate a single-dimensional regression feature quantity that is suitable for life prediction. The output layer is a single-neuron linear output structure that directly outputs the single-dimensional regression feature quantity output by the second connection layer, namely the fatigue remaining life of the welded structure.

[0030] Furthermore, the nonlinear ultrasonic Lamb wave detection of the welded structure with unknown damage state in step S7 uses the same excitation frequency, wedge angle, and detection parameters as the detection process of the specimen in step S3.

[0031] The present invention has the following beneficial effects: (1) This invention utilizes the high sensitivity of nonlinear ultrasound to the microcrack breathing effect to achieve reliable detection of early and mid-stage microcrack damage in welded structures, breaking through the technical bottleneck of traditional linear ultrasound being insensitive to small defects.

[0032] (2) The present invention first uses the pulse inversion method to perform initial purification and enhancement of the second harmonic component, suppressing the fundamental interference and canceling the background noise brought by the linear ultrasonic signal from the source; then, through empirical mode decomposition (EMD), the effective harmonic component is adaptively separated from random clutter and spatter noise, and irrelevant interference signals are filtered out in layers. After two-stage progressive signal enhancement processing, the amplitude of the effective second harmonic signal is significantly improved, and the noise floor is greatly suppressed. Compared with the original ultrasonic detection signal, the signal-to-noise ratio is significantly improved, solving the problem of strong scattering attenuation caused by the non-uniformity of the weld area and the coarse grain structure, making it easier to extract the nonlinear second harmonic component.

[0033] (3) This invention combines nonlinear ultrasonic testing results with fatigue loading data to establish a quantitative relationship between the microcrack damage evolution of welded structures and nonlinear ultrasonic response, so that the testing results can not only reflect the current damage state, but also be used for fatigue remaining life prediction.

[0034] (4) By introducing an artificial neural network model, this invention achieves effective modeling of complex nonlinear relationships, improves the accuracy and stability of welded structure life prediction, and reduces the reliance on human experience judgment.

[0035] (5) The nonlinear ultrasonic-based method for assessing microcrack damage and predicting life of welded structures proposed in this invention can achieve highly sensitive detection and life prediction in the early stage of micro-damage in welded structures. It has good engineering applicability and promotion value, and is of great significance for improving the safe operation level of welded structures of key equipment such as nuclear power and aerospace. Attached Figure Description

[0036] Figure 1 This is a flowchart of the damage assessment and life prediction process of this invention.

[0037] Figure 2 This is a numerical calculation graph of the dispersion curve in Example 1.

[0038] Figure 3 This is a set of signal waveforms acquired and processed by pulse inversion synthesis in Example 1.

[0039] Figure 4 It is a time-frequency diagram obtained by performing a short-time Fourier transform on the enhanced pulse inversion synthesized signal in Example 1. Detailed Implementation

[0040] The technical solution of the present invention will be further described in detail below with reference to specific embodiments. However, these embodiments are not intended to limit the present invention. Any similar structures and similar variations of the present invention should be included in the protection scope of the present invention. The commas in the present invention all indicate the relationship between and. The English letters in the present invention are case-sensitive.

[0041] like Figure 1 As shown, this invention provides a method for assessing microcrack damage and predicting lifespan of welded structures based on nonlinear ultrasound. The method includes: S1. Based on the material type, structural thickness and known acoustic properties of the welded structure, calculate the dispersion curve of ultrasonic Lamb wave in the welded structure, and obtain the distribution law of phase velocity and group velocity of different order Lamb wave modes with frequency. The specific steps for calculating the dispersion curve of ultrasonic Lamb waves in welded structures are as follows: Assume the Lamb wave propagates horizontally along the plate surface, and the phase velocity of the Lamb wave is... The following text is abbreviated as Frequency-thickness product range (MHz·mm); Define the normalized wavenumber component of the transverse wave in the thickness direction (perpendicular to the propagation direction). for:

[0042] Define the normalized wavenumber component of the longitudinal wave in the thickness direction (perpendicular to the propagation direction). for:

[0043] In the formula, The transverse wave velocity; For longitudinal wave velocity; Characteristic equations of the symmetric mode (S-mode):

[0044] Characteristic equation of antisymmetric mode (A-mode):

[0045] For each frequency-thickness product In the phase velocity range Internal scan, searching and The roots are then precisely determined using numerical methods such as the bisection method to obtain the symmetric modes of each order. ) and antisymmetric mode ( phase velocity ; Calculating group velocity from phase velocity using dispersion relation :

[0046] Finally, based on the known plate thickness The curves of phase velocity / group velocity as a function of frequency were obtained.

[0047] S2, based on the dispersion curve, select a suitable Lamb wave mode pair (high separation, stable propagation, and sensitive to microcrack damage) within a preset detection frequency range, and determine the center frequency, phase velocity, and wedge incident angle required to excite the mode pair. wedge incident angle The formula for calculation is:

[0048] In the formula, This indicates the longitudinal wave velocity within the wedge.

[0049] S3. Welded structural specimens with different microcrack damage levels were prepared by fatigue loading. Lamb waves were excited by an ultrasonic guided wave detection system through a determined wedge incident angle. The specimens with different damage states were detected by the pulse inversion method and the pulse inversion synthesized signal was collected. Among them, the fatigue loading adopted a constant amplitude cyclic loading method, and the damage state of the welded structure corresponding to each loading cycle was recorded.

[0050] The specific steps for detecting specimens in different damage states and acquiring raw ultrasonic guided wave signals using the pulse inversion method are as follows: For the same detection location, two sets of Lamb wave excitation signals with a phase difference of 180° were applied sequentially, and the corresponding time-domain response signals were acquired respectively. and :

[0051] in, For the amplitude of the excitation signal, The fundamental angular frequency, For time; The two sets of time-domain response signals are superimposed to obtain the pulse inversion composite signal. :

[0052] According to the nonlinear ultrasound propagation theory, pulse inversion synthesized signal The fundamental frequency and odd harmonics are effectively suppressed, while the second harmonic component is enhanced.

[0053] S4. Ultrasonic signal enhancement is performed on the pulse-inversion synthesized signal through empirical mode decomposition; then, a short-time Fourier transform is performed on the enhanced pulse-inversion synthesized signal to obtain the signal time-frequency diagram, and the amplitude of the corresponding frequency band is extracted to characterize the fundamental frequency component. Second harmonic component The nonlinear ultrasonic response parameters were calculated based on the fundamental frequency component and the second harmonic component. Due to the non-uniform microstructure and strong scattering attenuation caused by the coarse-grained structure in the weld region, the nonlinear second harmonic component in the directly acquired pulse inversion synthesized signal is extremely weak and difficult to extract effectively. Therefore, this invention first performs empirical mode decomposition enhancement processing on the acquired pulse inversion synthesized signal before the short-time Fourier transform (STFT) to effectively improve the second harmonic signal-to-noise ratio.

[0054] Based on the time-scale characteristics of the signal itself, the pulse is inverted to synthesize the signal. Adaptively decomposed into several intrinsic mode function (IMF) components With a residual component :

[0055] Specifically, the process involves first identifying the pulse inversion synthesized signal. Construct the upper envelope for each of the local maxima and minima. and lower envelope Calculate the local mean ; Then synthesize the signal from the pulse inversion. Subtract local mean from middle get ; Repeat the above process until Satisfying intrinsic mode function (IMF) components The two conditions to be met are: (1) the number of extreme points and the number of zero crossings are equal or differ by at most one over the entire signal length; (2) at any time, the mean of the upper envelope defined by the local maxima and the lower envelope defined by the local minima is zero.

[0056] That is, to obtain the first intrinsic mode function (IMF) component. ;use As a new signal to be decomposed, continue decomposition until the residual components are obtained. It is a monotonic function or a constant; The intrinsic mode function (IMF) components obtained from the decomposition Calculate its relationship with the theoretical second harmonic reference signal. Normalized correlation coefficient between :

[0057] Selecting the normalized correlation coefficient Largest intrinsic mode function (IMF) component As an effective component containing the second harmonic principal energy, the intrinsic mode function (IMF) component... The enhanced second harmonic signal is obtained by reconstructing the low-frequency trend term in the residual component. : .

[0058] Furthermore, in step S4, the enhanced second harmonic signal... Perform a short-time Fourier transform to obtain a time-frequency graph, and extract the amplitude of the corresponding frequency band to characterize the fundamental frequency component. Second harmonic component And based on the fundamental frequency component With second harmonic components Calculation of nonlinear ultrasonic response parameters : .

[0059] S5. Construct a dataset of "damage state-nonlinear ultrasonic parameters-remaining life" by combining the fatigue loading cycles and nonlinear ultrasonic response parameters corresponding to welded structural specimens with different crack damage states, and then normalize the dataset. Here, normalization refers to processing the nonlinear ultrasonic parameters and setting them to a range of 0 to 1. The normalization is performed in a conventional manner.

[0060] S6. Construct an artificial neural network model, using fatigue cycles and normalized nonlinear ultrasonic response parameters as model inputs, and the remaining fatigue life of the welded structure as model output. Train and optimize the artificial neural network model using the normalized dataset. The artificial neural network model is a feedforward neural network, a convolutional neural network, or a combination thereof, including an input layer, a first LSTM layer, a second LSTM layer, a first connection layer, a second connection layer, and an output layer connected in sequence. Each layer is connected to the previous one, and data is passed layer by layer in a unidirectional order. The input layer includes a multi-channel temporal input unit group, with each feature corresponding to an independent temporal input channel. The fatigue cycles and normalized nonlinear ultrasonic response parameters are recombined by the multi-channel temporal input unit to generate a temporal tensor. The first LSTM layer consists of multiple stacked LSTM memory units. Each memory unit integrates four core gating structures: a forget gate, an input gate, an output gate, and a cell state. The gating units uniformly adopt the Sigmoid activation function, and the cell state update is combined with the Tanh activation function to complete the nonlinear transformation. The temporal tensor is extracted by the gating structure to obtain a shallow temporal feature tensor, which is used to capture the long-distance temporal dependencies of the entire fatigue loading process, extract the basic temporal change law of the micro-damage evolution of the component, filter and monitor random interference, and retain the basic temporal information of fatigue damage evolution with each cycle. The second LSTM layer consists of multiple stacked LSTM memory units. Each memory unit integrates four core gating structures: a forget gate, an input gate, an output gate, and a cell state. The gating units uniformly adopt the Sigmoid activation function, and the cell state update is combined with the Tanh activation function to complete the nonlinear transformation. The shallow temporal feature tensor undergoes multi-layer gating iterative operations to complete deep feature abstraction based on the shallow temporal information, explore the high-order coupling correlation between fatigue cycles and nonlinear ultrasound multi-parameters, filter redundant temporal information, and output the aggregated high-dimensional deep feature tensor. The first connection layer is a standard fully connected network structure, consisting of neuron weight matrices, bias terms, and ReLU nonlinear activation functions. The high-dimensional feature tensor first undergoes linear weighting operations, and then a nonlinear transformation is achieved through the activation function. Simultaneously, the high-dimensional features are compressed and redundant features are removed, outputting a low-dimensional fused feature vector. This initially establishes a nonlinear mapping relationship between temporal damage features and welding fatigue remaining life. The second connection layer is a fully connected network structure. The low-dimensional fused feature vectors are subjected to linear weighting and nonlinear fitting again to further complete the feature dimension reduction, correct the previous mapping bias, refine and optimize the regression mapping curve from damage features to fatigue life, and generate a single-dimensional regression feature quantity that is suitable for life prediction. The output layer is a single-neuron linear output structure that directly outputs the single-dimensional regression feature quantity output by the second connection layer, namely the fatigue remaining life of the welded structure.

[0061] S7. Nonlinear ultrasonic Lamb wave detection is performed on the welded structure with unknown damage state to extract its nonlinear ultrasonic response parameters. The fatigue cycles and normalized nonlinear ultrasonic response parameters are input into the trained artificial neural network model, and the artificial neural network model outputs the fatigue remaining life, realizing the quantitative assessment of the microcrack damage degree of the welded structure and the prediction of the remaining life. The fatigue cycles of the welded structure with unknown damage state are obtained by the running time recorded by the fatigue testing machine.

[0062] In this process, the nonlinear ultrasonic Lamb wave detection of the welded structure with unknown damage state uses the same excitation frequency, wedge angle, and detection parameters as the detection process of the specimen in step S3.

[0063] Example 1 This embodiment uses a 4mm thick plate-shaped 6061 aluminum plate welded structure for specific illustration. This plate-shaped welded structure can be derived from welded components in high-reliability engineering equipment, such as aluminum alloy or other metal welded structures in aerospace equipment or nuclear power equipment. The welded structure uses friction stir welding to form the weld joint. During service, the weld area is prone to early microcrack damage under alternating loads. The specific steps are as follows: Based on the material type, thickness, and known acoustic properties of the welded structure, the dispersion curve of the ultrasonic Lamb wave in the welded structure is calculated, such as... Figure 2 As shown in the figure, for this 4mm thick 6061 aluminum plate, the S0 mode is relatively pure at approximately 600kHz without interference from the S1 mode. Although some A modes exist, their phase velocities differ significantly from those of S0, making them easy to distinguish. Therefore, 300kHz was chosen as the Lamb wave excitation frequency, and 600kHz as the second harmonic (nonlinear ultrasound) receiving frequency.

[0064] Based on the S0 mode pair, determine the corresponding center frequency, phase velocity, and wedge incident angle required to excite the mode. ; Welded structural specimens with different microcrack damage levels were prepared by fatigue loading. Lamb waves were excited by an ultrasonic guided wave detection system through a determined wedge incident angle. The specimens with different damage states were detected by the pulse inversion method and the pulse inversion synthesized signal was collected. One set of signal waveforms collected and processed by pulse inversion synthesis is shown in Figure 3. The welded structure was subjected to multiple tests under different fatigue loading cycles to obtain ultrasonic signal data corresponding to different damage states.

[0065] During fatigue loading, microcrack damage gradually forms and evolves within the welded structure. Under ultrasonic alternating stress, these microcracks exhibit periodic contact and separation behavior, known as the microcrack breathing effect, which introduces a significant nonlinear response during ultrasonic propagation.

[0066] Ultrasonic signal enhancement was performed on the pulse-inversion synthesized signal using empirical mode decomposition; and the time-frequency diagram of the enhanced pulse-inversion synthesized signal was obtained by performing a short-time Fourier transform. Figure 4 As shown, for time and frequency Figure 3 Amplitude retrieval is performed on the 3D matrix data, searching for the maximum amplitude within the range of 250 kHz to 350 kHz (implemented in code), and this maximum amplitude is denoted as... The maximum amplitude is searched within the range of 550 kHz to 650 kHz, and this maximum value is denoted as... ,use The nonlinear ultrasonic parameters were calculated. A dataset of "damage state-nonlinear ultrasonic parameters-remaining life" was constructed by comparing the fatigue loading cycles and nonlinear ultrasonic response parameters of welded structural specimens with different crack damage states. The dataset was then normalized and divided into training, validation, and test sets for the training and validation of the artificial neural network model.

[0067] Once the model is trained, nonlinear ultrasonic testing under the same conditions can be performed on welded structures with unknown damage states, followed by fatigue testing. The fatigue cycles obtained from the running time recorded by the fatigue testing machine and the extracted nonlinear ultrasonic response parameters are normalized and then input into the trained artificial neural network model. This enables quantitative assessment of the degree of microcrack damage in the welded structure and prediction of its remaining fatigue life. Notably, the degree of microcrack damage in the welded structure is quantitatively characterized by nonlinear ultrasonic parameters, rather than being directly output by the neural network.

[0068] This invention first employs a pulse inversion method to initially purify and enhance the second harmonic component, suppressing fundamental interference and canceling background noise from the linear ultrasonic signal at its source. Then, empirical mode decomposition (EMD) adaptively separates the effective harmonic component from random clutter and spatter noise, filtering out irrelevant interference signals in layers. Through this two-stage progressive signal enhancement process, the amplitude of the effective second harmonic signal is significantly increased, and the noise floor is greatly suppressed. Compared to the original ultrasonic detection signal, the signal-to-noise ratio is significantly improved, resolving the problem of strong scattering attenuation caused by the non-uniform microstructure and coarse-grained structure in the weld region, making it easier to extract the nonlinear second harmonic component. By combining the nonlinear ultrasonic detection results with fatigue loading data, a quantitative relationship between the microcrack damage evolution of the welded structure and the nonlinear ultrasonic response is established, enabling the detection results to not only reflect the current damage state but also be used for fatigue remaining life prediction.

[0069] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

Claims

1. A method for assessing microcrack damage and predicting lifespan of welded structures based on nonlinear ultrasound, characterized in that, The method includes: S1. Based on the material type, structural thickness and known acoustic properties of the welded structure, calculate the dispersion curve of ultrasonic Lamb wave in the welded structure, and obtain the distribution law of phase velocity and group velocity of different order Lamb wave modes with frequency. S2, based on the dispersion curve, select a suitable Lamb wave mode pair within the preset detection frequency range, and determine the center frequency, phase velocity, and wedge incident angle required to excite the mode corresponding to the mode pair; S3, welded structural specimens with different microcrack damage levels were prepared by fatigue loading. Lamb waves were excited by an ultrasonic guided wave detection system through a determined wedge incident angle. The pulse inversion method was used to detect the specimens with different damage states and collect the pulse inversion synthesized signal. S4. The pulse inversion synthesized signal is enhanced with ultrasound signal by empirical mode decomposition; and the enhanced pulse inversion synthesized signal is subjected to short-time Fourier transform to obtain time-frequency diagram, and the amplitude of the corresponding frequency band is extracted to characterize the fundamental frequency component and the second harmonic component, and the nonlinear ultrasound response parameters are calculated based on the fundamental frequency component and the second harmonic component. Due to the non-uniform microstructure and strong scattering attenuation caused by the coarse-grained structure in the weld area, the nonlinear second harmonic component in the directly acquired pulse inversion synthesized signal is extremely weak and difficult to extract effectively. Therefore, before the short-time Fourier transform, the acquired pulse inversion synthesized signal is first subjected to empirical mode decomposition enhancement processing to effectively improve the second harmonic signal-to-noise ratio. Based on the time-scale characteristics of the signal itself, the pulse is inverted to synthesize the signal. Adaptively decomposed into several intrinsic mode function components With a residual component : Specifically, the process involves first identifying the pulse inversion synthesized signal. Construct the upper envelope for each of the local maxima and minima. and lower envelope Calculate the local mean ; Then synthesize the signal from the pulse inversion. Subtract local mean from middle get ; Repeat the above process until Satisfying intrinsic mode function components The following two conditions must be met: 1) Over the entire signal length, the number of extreme points and the number of zero crossings are equal or at most differ by one; 2) At any given time, the mean of the upper envelope defined by the local maxima and the lower envelope defined by the local minima is zero. That is, to obtain the first intrinsic mode function component. ;use As a new signal to be decomposed, continue decomposition until the residual components are obtained. It is a monotonic function or a constant; The eigenmode function components obtained from the decomposition Calculate its relationship with the theoretical second harmonic reference signal. Normalized correlation coefficient between : Selecting the normalized correlation coefficient Largest eigenmode function component As an effective component containing the second harmonic principal energy, the intrinsic mode function (IMF) component... The enhanced second harmonic signal is obtained by reconstructing the low-frequency trend term in the residual component. : For the enhanced second harmonic signal Perform a short-time Fourier transform to obtain a time-frequency graph, and extract the amplitude of the corresponding frequency band to characterize the fundamental frequency component. Second harmonic component And based on the fundamental frequency component With second harmonic components Calculation of nonlinear ultrasonic response parameters : S5. The fatigue loading cycles and nonlinear ultrasonic response parameters corresponding to welded structural specimens with different crack damage states are used to construct a dataset of "damage state-nonlinear ultrasonic parameters-remaining life", and the dataset is normalized. S6. Construct an artificial neural network model, using fatigue cycles and normalized nonlinear ultrasonic response parameters as model inputs, and the remaining fatigue life of the welded structure as model output. Train and optimize the artificial neural network model using the normalized dataset. S7 performs nonlinear ultrasonic Lamb wave detection on welded structures with unknown damage states, extracts their nonlinear ultrasonic response parameters, inputs the fatigue cycles and normalized nonlinear ultrasonic response parameters into a trained artificial neural network model, and outputs the fatigue remaining life from the artificial neural network model, thereby realizing the quantitative assessment of the degree of microcrack damage and the prediction of the remaining life of the welded structure.

2. The method for assessing microcrack damage and predicting lifespan of welded structures based on nonlinear ultrasound according to claim 1, characterized in that, The specific steps for calculating the dispersion curve of the ultrasonic Lamb wave in the welded structure in step S1 are as follows: Assume the Lamb wave propagates horizontally along the plate surface, and the phase velocity of the Lamb wave is... The following text is abbreviated as Frequency-thickness product range ; Define the normalized wavenumber component of the transverse wave in the thickness direction of the plate. for: Define the normalized wavenumber components of the longitudinal wave in the thickness direction of the plate. for: In the formula, The transverse wave velocity; For longitudinal wave velocity; Symmetric mode characteristic equation: Antisymmetric mode characteristic equation: For each frequency-thickness product In the phase velocity range Internal scan, searching and The roots are then determined using numerical methods such as the bisection method to obtain the precise roots, yielding the phase velocities of each order of symmetric and antisymmetric modes. ; Calculating group velocity from phase velocity using dispersion relation : Finally, based on the known plate thickness The curves of phase velocity / group velocity as a function of frequency were obtained.

3. The method for assessing microcrack damage and predicting lifespan of welded structures based on nonlinear ultrasound according to claim 1, characterized in that, The angle of incidence of the wedge in step S2 The formula for calculation is: In the formula, This indicates the longitudinal wave velocity within the wedge.

4. The method for assessing microcrack damage and predicting lifespan of welded structures based on nonlinear ultrasound according to claim 1, characterized in that, The specific steps in step S3 for detecting specimens in different damage states and acquiring the original ultrasonic guided wave signals using the pulse inversion method are as follows: For the same detection location, two sets of Lamb wave excitation signals with a phase difference of 180° were applied sequentially, and the corresponding time-domain response signals were acquired respectively. and : in, For the amplitude of the excitation signal, The fundamental angular frequency, For time; The two sets of time-domain response signals are superimposed to obtain the pulse inversion composite signal. : According to the nonlinear ultrasound propagation theory, pulse inversion synthesized signal The fundamental frequency and odd harmonics are effectively suppressed, while the second harmonic component is enhanced.

5. The method for assessing microcrack damage and predicting lifespan of welded structures based on nonlinear ultrasound according to claim 1, characterized in that, In step S6, the artificial neural network model is a feedforward neural network, a convolutional neural network, or a combination thereof, including an input layer, a first LSTM layer, a second LSTM layer, a first connection layer, a second connection layer, and an output layer connected in sequence. Each layer is connected to the previous one, and data is passed layer by layer in a unidirectional order. The input layer includes a multi-channel temporal input unit group, with each feature corresponding to an independent temporal input channel. The fatigue cycles and normalized nonlinear ultrasonic response parameters are recombined by the multi-channel temporal input unit to generate a temporal tensor. The first LSTM layer consists of multiple stacked LSTM memory units. Each memory unit integrates four core gating structures: a forget gate, an input gate, an output gate, and a cell state. The gating units uniformly adopt the Sigmoid activation function, and the cell state update is combined with the Tanh activation function to complete the nonlinear transformation. After the temporal tensor is extracted by the gated structure, a shallow temporal feature tensor is obtained, which is used to capture the long-distance temporal dependency relationship of the entire fatigue loading process, extract the basic temporal change law of the micro-damage evolution of the component, filter out random interference, and retain the basic temporal information of fatigue damage evolution with each cycle. The second LSTM layer consists of multiple stacked LSTM memory units. Each memory unit integrates four core gating structures: a forget gate, an input gate, an output gate, and a cell state. The gating units uniformly adopt the Sigmoid activation function, and the cell state update is combined with the Tanh activation function to complete the nonlinear transformation. The shallow temporal feature tensor undergoes multi-layer gating iterative operations to complete deep feature abstraction based on the shallow temporal information, explore the high-order coupling correlation between fatigue cycles and nonlinear ultrasound multi-parameters, filter redundant temporal information, and output the aggregated high-dimensional deep feature tensor. The first connection layer is a standard fully connected network structure, consisting of neuron weight matrices, bias terms, and ReLU nonlinear activation functions. The high-dimensional feature tensor first undergoes linear weighting operations, and then a nonlinear transformation is achieved through the activation function. Simultaneously, the high-dimensional features are compressed and redundant features are removed, outputting a low-dimensional fused feature vector. This initially establishes a nonlinear mapping relationship between temporal damage features and welding fatigue remaining life. The second connection layer is a fully connected network structure. The low-dimensional fused feature vectors are subjected to linear weighting and nonlinear fitting again to further complete the feature dimension reduction, correct the previous mapping bias, refine and optimize the regression mapping curve from damage features to fatigue life, and generate a single-dimensional regression feature quantity that is suitable for life prediction. The output layer is a single-neuron linear output structure that directly outputs the single-dimensional regression feature quantity output by the second connection layer, namely the fatigue remaining life of the welded structure.

6. The method for assessing microcrack damage and predicting lifespan of welded structures based on nonlinear ultrasound according to claim 1, characterized in that, In step S7, the nonlinear ultrasonic Lamb wave detection of the welded structure with unknown damage state uses the same excitation frequency, wedge angle, and detection parameters as the detection process of the specimen in step S3.

Images