Ultrasonic guided wave mixing for circumferential localization of delaminations in filament wound tubular structures

Abstract

This invention provides an ultrasonic guided wave mixing circumferential localization method for delamination damage in fiber-wound tubular structures. By employing mixing excitation, a mixing region is formed within the fiber-wound tubular structure. For delamination damage caused by low-energy impacts that cannot be detected or accurately located using conventional methods, the method utilizes the local contact acoustic nonlinearity. When the mixing region coincides with the delamination damage, difference frequency components and sum frequency components are generated. Therefore, relying on the mixing components that satisfy the matching criterion, accurate detection and localization of such delamination damage can be achieved, including axial and circumferential localization. Furthermore, by using a deep learning model to predict the circumferential angle of the delamination damage and combining it with an angle increment discrimination method to determine the true circumferential angle, "end-to-end" circumferential angle prediction of the delamination damage is achieved, avoiding human error and improving the accuracy and efficiency of non-destructive testing and localization.

Description

Technical Field

[0001] This invention relates to the field of non-destructive testing technology for pipelines, specifically to an ultrasonic guided wave mixing circumferential localization method for layered damage in fiber-wound tubular structures. Background Technology

[0002] Pipeline transportation, as one of the five major modes of transportation, boasts advantages such as safety, environmental friendliness, low energy consumption, and minimal losses. Achieving ultra-high pressure, ultra-large flow, and ultra-long-distance transmission is the development trend of buried pipeline transportation, which also places higher demands on pipeline material properties. Currently, pipeline steel has a complete API (American Petroleum Institute) standard, but the problems of poor corrosion resistance and crack arrest ability of pipeline steel remain unresolved; once a crack occurs in the pipeline, the crack will propagate for several kilometers. Compared with pipeline steel, composite material pipes have lower density, higher specific strength, and stronger corrosion resistance, but poor fluid sealing and are prone to deformation. To address these issues, researchers have developed glass fiber wound steel pipes by winding continuous glass fibers at a specific angle onto pipeline steel, thus overcoming the shortcomings of both pipeline steel and composite material pipes. However, fiberglass-wound steel pipes also have their problems: The impact of falling rocks caused by landslides is one of the most serious safety threats to buried pipelines. Even with insufficient impact energy, although the pipeline may not undergo significant plastic deformation, localized plastic deformation of the polycrystalline metal occurs at the impact site. The fiberglass reinforced plastic (GFRP) matrix and fibers break, and delamination occurs within the GFRP and between the GFRP and the polycrystalline metal. These microstructural changes will continue to evolve over time, ultimately leading to the failure of the fiberglass-wound steel pipe and accidents.

[0003] Because low-energy impact damage in GFRP differs significantly from damage in polycrystalline metals, traditional non-destructive testing (NDT) methods such as eddy current, magnetic particle, and visual inspection are no longer applicable. Currently, NDT methods based on infrared, X-ray, fiber optic, acoustic emission, and ultrasonic waves are gaining increasing attention. Compared to the aforementioned methods, the second harmonic and nonlinear ultrasonic modulation of ultrasonic guided waves can scan a wide area of ​​the sample and detect microstructural changes, but it cannot pinpoint the location of defects / damage.

[0004] In summary, for fiber-wound tubular structures such as fiberglass-wound steel pipes, existing ultrasonic testing methods are unable to effectively detect and determine the degree and location of low-energy impact damage, affecting the maintenance of such pipes. Therefore, there is an urgent need for a new ultrasonic testing and localization method for low-energy impact damage in such pipes. Summary of the Invention

[0005] This invention addresses the aforementioned problems by providing an ultrasonic detection and localization method for fiber-wound tubular structures such as fiberglass-wound steel pipes, capable of effectively detecting and determining the degree and location of low-energy impact damage. To address these issues, the inventors introduce ultrasonic guided wave counter-mixing technology for damage detection in fiber-wound tubular structures, enabling axial and circumferential localization of layered damage. This allows for precise maintenance of buried pipelines to prevent accidents. The invention employs the following technical solution:

[0006] This invention provides an ultrasonic guided wave mixing circumferential localization method for delamination damage in a fiber-wound tubular structure. The method comprises the following steps: Step S1, acquiring the material and geometric parameters of the fiber-wound tubular structure, and applying mixing excitation to the structure using a nonlinear ultrasonic measurement system based on these parameters to form a mixing region within the structure; Step S2, adjusting the axial position of the mixing region within the structure by changing the parameters of the mixing excitation, and axially locating the delamination damage based on the mixing components; Step S3, placing the mixing region at the location of the delamination damage based on the axial positioning result, and circumferentially locating the delamination damage based on the mixing components. In Step S3, a circumferential localization prediction model is used to obtain multiple predicted values ​​of the circumferential angle of the delamination damage based on the mixing components, and an angle increment discrimination method is used to determine the true circumferential angle among the multiple predicted values, thereby achieving the circumferential localization.

[0007] The ultrasonic guided wave mixing circumferential localization method for layered damage of fiber-wound tubular structures provided by the present invention may also have the following technical features, wherein step S1 includes the following sub-steps: step S1-1, obtaining the material parameters and geometric parameters of the fiber-wound tubular structure; step S1-2, screening ultrasonic guided wave opposing mixing mode pairs based on the material parameters and geometric parameters, and arranging the receiving point positions based on the screened mode pairs; step S1-3, applying two ultrasonic guided wave signals with different center frequencies and opposite directions to both ends of the fiber-wound tubular structure using the nonlinear ultrasonic measurement system to form the mixing region.

[0008] The ultrasonic guided wave mixing circumferential localization method for delamination damage in fiber-wound tubular structures provided by this invention may also have the following technical features, wherein step S2 includes the following sub-steps: Step S2-1, by changing the time delay between the two ultrasonic guided wave signals, the position of the mixing region is adjusted along the axial direction of the fiber-wound tubular structure to achieve axial scanning; Step S2-2, during the axial scanning process, response signals from multiple different receiving points along the axial direction are acquired, and the normalized nonlinear acoustic energy at each intersection position of the two ultrasonic guided wave signals is calculated based on the multiple response signals; Step S2-3, a relationship curve between the intersection position and the normalized nonlinear acoustic energy is generated; Step S2-4, based on the relationship curve and the functional relationship between the delamination damage and the normalized nonlinear acoustic energy, the axial position of the delamination damage is determined.

[0009] The ultrasonic guided wave mixing circumferential localization method for layered damage of fiber-wound tubular structures provided by this invention may also have the following technical features: in step S1-2, a non-dispersive torsional mode T(0,1) is selected. In step S1-3, a center frequency f is applied to one end of the fiber-wound tubular structure. a The number of cycles is n a The fundamental wave a, with a center frequency f applied at its other end b The number of cycles is n b The fundamental wave b, and both fundamental waves a and b are ultrasonic guided waves, with receiving points respectively set at both ends of the fiber-wound tubular structure. In step S2-1, a time delay t is created between fundamental wave a and fundamental wave b. d The intersection location is represented as:

[0010]

[0011] In the formula, l is the length of the fiber-wound tubular structure. , These are the group velocities of fundamental wave a and fundamental wave b, respectively. In step S2-2, the original time-domain signal is extracted from the response signal and subjected to phase inversion processing. A Fourier transform is performed on the phase-inverted original time-domain signal to obtain its spectrum. The amplitudes of the difference frequency and all frequencies near the sum frequency are extracted within a specific bandwidth of this spectrum using a gate function. The integral of the squared amplitudes is then calculated as the normalized nonlinear acoustic energy.

[0012]

[0013]

[0014] In the formula, A(f) is the spectrum of the time-domain signal after phase reversal processing, G(f) is the gate function, and flow and f high These represent the lower and upper frequency limits for a specific bandwidth. In steps S2-4, based on the relationship curve, the functional relationship between the delamination damage and the normalized nonlinear acoustic energy, the z-coordinate of the delamination damage is obtained, wherein the z-axis coincides with the axial direction of the fiber-wound tubular structure.

[0015] The ultrasonic guided wave mixing circumferential localization method for layered damage of fiber-wound tubular structures provided by the present invention may also have the following technical feature: in steps S1-3, the difference frequency and sum frequency of the center frequencies of the fundamental wave a and the fundamental wave b avoid the center frequency f. a Or the center frequency f b The frequency harmonics. In step S2-1, the time delay is applied only to one of the fundamental frequency a and the fundamental frequency b, and in a specific step size. An axial scan is performed on the fiber-wound tubular structure, wherein c g Let Δt be the group velocity of the fundamental wave a and the fundamental wave b. d This represents the change in time delay.

[0016] The ultrasonic guided wave mixing circumferential localization method for layered damage of fiber-wound tubular structures provided by the present invention may also have the following technical features, wherein, in step S2-2, the phases of the fundamental wave a and the fundamental wave b are set to 0° and 0°, 0° and 180°, 180° and 0°, and 180° and 180°, respectively, and the corresponding original time-domain signals S are acquired respectively. a,b (t), S a,-b (t), S -a,b (t) and S -a,-b (t), the original time-domain signal after phase inversion processing is expressed as:

[0017] .

[0018] The ultrasonic guided wave mixing circumferential localization method for layered damage of fiber-wound tubular structures provided by the present invention may also have the following technical features, wherein step S3 includes the following sub-steps: Step S3-1, placing the mixing region at the axial position of the layered damage, and for each receiving point, collecting response signals from multiple different receiving points in the circumferential direction, and constructing multiple datasets based on the multiple sets of response signals; Step S3-2, inputting the multiple datasets into the circumferential localization prediction model to obtain multiple predicted values ​​of the circumferential angle of the layered damage, wherein the circumferential localization prediction model is a trained one-dimensional convolutional neural network model; Step S3-3, using the angle increment discrimination method to determine the true circumferential angle of the layered damage based on the multiple predicted values, thereby determining the circumferential position of the layered damage.

[0019] The ultrasonic guided wave mixing circumferential localization method for layered damage in fiber-wound tubular structures provided by this invention also has the following technical features: In step S3-1, the time delay is determined based on the axial localization result, thereby adjusting the position of the mixing region so that the intersection position coincides with the center of the layered damage in the axial direction. Two sets of time-domain signals are collected at the first and second receiving points in the circumferential direction, respectively. After preprocessing the two sets of time-domain signals, corresponding first and second datasets are obtained. In step S3-2, the first and second datasets are input into the circumferential localization prediction model to obtain the first and second predicted values ​​of the circumferential angle. In step S3-3, the true φ coordinates of the layered damage are determined based on the first and second predicted values ​​using the angle increment discrimination method.

[0020]

[0021] In the formula, The circumferential angle between the second receiving point and the delamination damage. The first predicted value, The second predicted value is ∆φ, where ∆φ is a predetermined small angular increment, φ is the rotation angle of the r-axis, and the r-axis is aligned with the radial direction of the fiber-wound tubular structure.

[0022] The ultrasonic guided wave mixing circumferential localization method for layered damage in fiber-wound tubular structures provided by this invention also has the following technical features: In step S3-1, the constructed dataset includes dual-channel data; in step S3-2, the one-dimensional convolutional neural network model has a dual-channel input module, a feature extraction module, an adaptive average pooling module, and a multi-task module. The multi-task module includes an angle prediction branch and a waveform reconstruction branch. During model training, the dataset is input into the one-dimensional convolutional neural network model. The predicted value is output through the angle prediction branch, and the waveform is reconstructed through the waveform reconstruction branch to capture the mixing components with cumulative effects. Five-fold cross-validation is used, and the weights and parameters of the convolutional layers are iteratively updated based on a loss function using the Adam optimizer. The loss function is expressed as:

[0023]

[0024] In the formula, L angle For the mean square error of angle prediction, L wave The mean square error for waveform reconstruction. is the regularization term, and s1 and s2 are the learnable parameters for the angle prediction task and the waveform reconstruction task, respectively, and are homoscedastic uncertainty parameters.

[0025] The ultrasonic guided wave mixing circumferential localization method for delamination damage of fiber-wound tubular structures provided by the present invention may also have the following technical features: For the dual-channel data, based on the wavenumbers of the fundamental wave a and the fundamental wave b, the mode of the generated mixing component is determined by using a phase matching criterion; the corresponding group velocity is determined based on the mixing component; the size of the time window is set based on the group velocity; the time window is used to select the time-domain signal containing the mixing component sensitive to the delamination damage in the original time-domain signal; after processing the time-domain signal using the phase inversion method, the amplitude of all frequencies near the difference frequency or sum frequency is extracted within a specific bandwidth; and the amplitude is normalized to obtain multiple data points.

[0026] The role and effect of invention

[0027] The ultrasonic guided wave mixing circumferential localization method for delamination damage in fiber-wound tubular structures provided by this invention utilizes mixing excitation to form a mixing region within the fiber-wound tubular structure. For delamination damage caused by low-energy impacts that cannot be detected or accurately located using conventional methods, the method leverages the local contact acoustic nonlinearity to generate mixing components when the mixing region coincides with the delamination damage. Therefore, it can rely on these mixing components to achieve accurate detection and localization of such delamination damage, including axial and circumferential localization. Furthermore, by employing a deep learning model to predict the circumferential angle of the delamination damage and combining it with an angle increment discrimination method to determine the true circumferential angle, it achieves end-to-end circumferential angle prediction of the delamination damage, avoiding human error and improving the efficiency of non-destructive testing and localization. Attached Figure Description

[0028] Figure 1 This is a flowchart of the ultrasonic guided wave mixing circumferential localization method for layered damage of fiber-wound tubular structures in an embodiment of the present invention;

[0029] Figure 2 This is a flowchart of the sub-steps of the ultrasonic guided wave mixing circumferential localization method for layered damage of fiber-wound tubular structures in an embodiment of the present invention;

[0030] Figure 3 This is a schematic diagram of the structure of a one-dimensional convolutional neural network model in an embodiment of the present invention;

[0031] Figure 4 This is a schematic diagram of a fiber-wound tubular structure with layered damage and an excitation signal in an embodiment of the present invention;

[0032] Figure 5 This is an example diagram of the original time-domain signal in an embodiment of the present invention;

[0033] Figure 6 This is an example diagram of the spectrum of the original time-domain signal in an embodiment of the present invention;

[0034] Figure 7 This is an example diagram of the time-domain signal after phase reversal processing in an embodiment of the present invention;

[0035] Figure 8 This is an example diagram of the spectrum of the time-domain signal after phase reversal processing in an embodiment of the present invention;

[0036] Figure 9 This is a graph showing the normalized nonlinear acoustic energy of the intersection location and the difference frequency component in an embodiment of the present invention.

[0037] Figure 10 This is a graph showing the normalized nonlinear acoustic energy of the intersection location and the sum-frequency component in an embodiment of the present invention.

[0038] Figure 11 This is a root mean square error diagram of the angle prediction of time-domain signals in different time windows in an embodiment of the present invention;

[0039] Figure 12 This is a fitting diagram of the actual angle and predicted angle of layered damage in an embodiment of the present invention;

[0040] Figure 13 This is a comparison chart of four evaluation indicators for four angle prediction methods in the embodiments of the present invention. Detailed Implementation

[0041] To make the technical means, creative features, objectives and effects of this invention easy to understand, the ultrasonic guided wave mixing circumferential positioning method for delamination damage of glass fiber wound steel pipe of this invention will be specifically described below with reference to embodiments and accompanying drawings.

[0042] Example

[0043] Figure 1 This is a flowchart of the ultrasonic guided wave mixing circumferential localization method for layered damage of fiber-wound tubular structures in this embodiment. Figure 2 This is a flowchart of the sub-steps of the ultrasonic guided wave mixing circumferential localization method for layered damage of fiber-wound tubular structures in this embodiment.

[0044] like Figure 1 As shown, the method includes the following steps:

[0045] In preparation step S1, the physical and geometric parameters of the fiber-wound tubular structure are obtained, and a nonlinear ultrasonic measurement system is used to apply a mixing excitation to the fiber-wound tubular structure to form a mixing region in the fiber-wound tubular structure.

[0046] In the axial positioning step S2, the position of the mixing zone in the axial direction of the fiber-wound tubular structure is adjusted by changing the parameters of the mixing excitation, and the delamination damage is axially located based on the mixing components.

[0047] In the circumferential positioning step S3, the mixing region is placed at the location of the layered damage based on the axial positioning result, and the layered damage is circumferentially positioned based on the mixing components. In step S3, the circumferential positioning prediction model is used to obtain multiple predicted values ​​of the circumferential angle of the layered damage based on the response signal, and the angle increment discrimination method is used to distinguish the true circumferential angle among the multiple predicted values, thereby realizing the circumferential positioning.

[0048] The steps described above will be explained in detail below.

[0049] In preparation step S1, the material and geometric parameters of the fiber-wound tubular structure are obtained, and based on these parameters, a nonlinear ultrasonic measurement system is used to apply a mixing excitation to the fiber-wound tubular structure to form a mixing region in the fiber-wound tubular structure.

[0050] The nonlinear ultrasonic measurement system includes an arbitrary function generator, a pulse amplifier, an excitation transducer, a receiving transducer, a filter, and an oscilloscope. The excitation transducer and the receiving transducer are piezoelectric ceramic array units.

[0051] like Figure 2 As shown, step S1 specifically includes the following sub-steps:

[0052] Step S1-1: Obtain the material and geometric parameters of the fiber-wound tubular structure.

[0053] In this step, the material parameters include the material properties of the various materials that make up the tubular structure, and the geometric parameters include at least the length and outer diameter of the tubular structure.

[0054] Steps S1-2 involve screening ultrasonic guided wave opposing mixing mode pairs based on the material parameters and dimensions of the fiber-wound tubular structure, and arranging the receiving points based on the screened mode pairs.

[0055] In this step, for example, the dispersion equation of the ultrasonic guided wave in the fiber-wound tubular structure can be established based on the material and geometric parameters of the fiber-wound tubular structure, and the dispersion curves of different modes can be plotted based on the dispersion equation. Then, appropriate mode pairs and excitation frequencies can be selected based on the dispersion curves of each mode, and the receiving point positions can be determined at both ends of the fiber-wound tubular structure based on the selected mode pairs. Receiving transducers (piezoelectric ceramic array units) are arranged at the receiving point positions.

[0056] In this embodiment, considering the dispersion and multimode nature of ultrasonic guided waves, the non-dispersion torsional mode T(0,1) is preferentially selected.

[0057] Steps S1-3: Using a nonlinear ultrasonic measurement system, two ultrasonic guided wave signals with different center frequencies are applied to both ends of the fiber-wound tubular structure to form the aforementioned mixing zone.

[0058] In this step, two ultrasonic guided waves with different center frequencies and opposite propagation directions are simultaneously applied by excitation transducers set at both ends of the fiber-wound tubular structure to excite counter-frequency mixing. The two ultrasonic guided waves propagate and meet in the fiber-wound tubular structure to form a mixing zone with a specific length.

[0059] Specifically, an arbitrary function generator generates a sinusoidal pulse signal with a Hanning window and a specific center frequency and number of periods. This sinusoidal pulse signal is amplified by a pulse amplifier, and then converted into mechanical vibration by a piezoelectric ceramic array unit and applied to the end of the fiber-wound tubular structure, thereby applying displacement along the circumference of the tubular structure. In this embodiment, a center frequency of f is applied to the left end of the fiber-wound tubular structure. a The number of cycles is n a The ultrasonic guided wave signal a (fundamental wave a) is applied at the right end, while a center frequency f is applied at the right end. b The number of cycles is n b The ultrasonic guided wave signal b (fundamental wave b).

[0060] In addition, to avoid interference from higher harmonics, the difference frequency (|f a -f b |) and sum frequency (|f) a +f b |) Avoid the center frequency f a or f b The frequency multiplier.

[0061] In the axial positioning step S2, the position of the mixing zone in the axial direction of the fiber-wound tubular structure is adjusted by changing the parameters of the mixing excitation, and the delamination damage is axially located based on the mixing components.

[0062] like Figure 2 As shown, step S2 specifically includes the following sub-steps:

[0063] Step S2-1: By changing the time delay of the two ultrasonic guided wave signals, the position of the mixing zone is adjusted along the axial direction of the fiber-wound tubular structure, thereby achieving axial scanning.

[0064] In this step, firstly, the spatial lengths of fundamental frequency a and fundamental frequency b are calculated:

[0065] (1)

[0066] In the formula, n a n af represents the number of periods for fundamental wave a and fundamental wave b, respectively. a f b These are the center frequencies of the two, , These are the phase velocities of the two, respectively.

[0067] Then, based on the characteristics of the opposing mixing of ultrasonic guided wave signals, the spatial length of the mixing region is obtained:

[0068] (2)

[0069] In the formula, , These are the group velocities of fundamental wave a and fundamental wave b, respectively.

[0070] The location Z where fundamental frequencies a and b intersect m With time delay t d In order to achieve axial scanning of the entire fiber-wound tubular structure and make it easier to calculate and control, in this embodiment, a time delay is applied only to the fundamental wave a at the left end or the fundamental wave b at the right end.

[0071] Assuming the z-coordinate of the piezoelectric ceramic array element used to generate the fundamental wave a is z=0, when t d When = 0, based on the group velocities of the two fundamental waves, their intersection position Z is... m The z-coordinate is When no time delay is applied to the fundamental frequency a, but only to the fundamental frequency b, a time delay t is applied... d At that time, the intersection point Z m Shift to the right; when no time delay is applied to the fundamental frequency b, but only to the fundamental frequency a. d At that time, the intersection point Z m Shift to the left, intersection point Z m The expression is:

[0072] (3)

[0073] In the formula, l is the length of the entire fiber-wound tubular structure. When t d When t > 0, the excitation time of the left-end fundamental frequency a is earlier than the excitation time of the right-end fundamental frequency b, and the mixing region shifts to the right. d When <0, the excitation time of the right-end fundamental wave b is earlier than the excitation time of the left-end fundamental wave a, and the term in formula (3) If the value is less than 0, the mixing region shifts to the left.

[0074] Therefore, in this step, only one variable needs to be adjusted, namely the time delay t. d This allows for axial scanning of the entire fiber-wound tubular structure.

[0075] Regarding time delay td Adjusting the formula (3), the slope is equal to a constant. In this embodiment, equal time intervals are used to adjust the time delay t. d This is to ensure that the mixing region moves to the left or right along the axial direction with equal steps. Specifically, both fundamental frequency a and fundamental frequency b are non-dispersive T(0,1) modes. = = ), Simplified to Therefore, in this embodiment, a specific step size is used. An axial scan was performed on the entire fiber-wound tubular structure, where c g Let Δt be the group velocity of fundamental wave a and fundamental wave b. d This represents the change in time delay.

[0076] Step S2-2: During the axial scanning process, the response signals of multiple different receiving points along the axial direction are acquired, and the normalized nonlinear acoustic energy of each intersection position of the two ultrasonic guided waves is obtained based on the multiple response signals.

[0077] When the mixing region coincides with the delamination damage, the contact acoustic nonlinearity of the delamination damage induces nonlinear effects, generating difference frequency components and sum frequency components. Therefore, ultrasonic guided wave mixing can be used to locate the delamination damage axially.

[0078] Specifically, the piezoelectric ceramic array unit converts the mechanical vibration of the receiving point into a response signal. The response signal is then processed by an oscilloscope or data acquisition card to extract the original time-domain signal S. a,b (t). To suppress the fundamental and second harmonics and highlight the difference frequency or sum frequency components, phase reversal technology is used in this embodiment. The arbitrary function generator sets the phases of the fundamental a and fundamental b to 0° and 0°, 0° and 180°, 180° and 0°, and 180° and 180° respectively. The original time-domain signals are acquired for these four sets of settings, and the corresponding original time-domain signals are denoted as S respectively. a,b (t), S a,-b (t), S -a,b (t) and S -a,-b (t). Then, the above response signals are linearly superimposed and divided by 4 to obtain the original time-domain signal after phase reversal processing, which highlights the difference frequency component and the sum frequency component:

[0079] (4)

[0080] A Fourier transform is performed on the original time-domain signal S(t) after phase reversal processing to obtain the corresponding spectrum A(f). Then, a suitable gate function G(f) is selected, and the amplitudes of the difference frequency and all frequencies near the sum frequency are extracted within a specific bandwidth of the spectrum A(f). The integral of the square of the amplitude within this specific bandwidth is calculated as the normalized nonlinear acoustic energy β. energy .

[0081] Normalized nonlinear acoustic energy β energy The gate function G(f) is expressed as follows:

[0082] (5)

[0083] (6)

[0084] In the formula, A(f) is the spectrum of the time-domain signal after phase reversal processing, G(f) is the gate function, and f low and f high These are the lower and upper frequency limits for a specific bandwidth, respectively.

[0085] Step S2-3 generates the relationship curve between the intersection location and the normalized nonlinear acoustic energy.

[0086] In this step, the normalized nonlinear acoustic energy β is used. energy Using the vertical axis as the coordinate and the intersection point Z as the coordinate... m Used as the x-axis to plot a curve.

[0087] Step S2-4: Based on the relationship curve, the functional relationship between the layered damage and the normalized nonlinear acoustic energy, determine the axial position of the layered damage.

[0088] When the mixing region coincides with the delamination damage, difference frequency components and sum frequency components appear. Furthermore, the closer the intersection of the two ultrasonic guided wave signals is to the center of the delamination damage, the higher the normalized nonlinear acoustic energy β of the difference frequency components and sum frequency components. energy The larger it is, the greater the normalized nonlinear acoustic energy β can be found in the relationship curve in this step. energy The peak value, and based on the intersection position Z corresponding to the peak value. m The z-coordinate of the layered damage is obtained from the coordinates.

[0089] In the circumferential positioning step S3, the mixing region is placed at the location of the delamination damage based on the axial positioning result, and the delamination damage is circumferentially located based on the response signal after mixing.

[0090] like Figure 2 As shown, step S3 specifically includes the following sub-steps:

[0091] Step S3-1: Place the mixing region at the axial position of the layered damage. For each receiving point, collect the response signals of multiple different receiving points in the circumferential direction, and construct multiple datasets based on the multiple sets of response signals.

[0092] In this step, the time delay between fundamental wave a and fundamental wave b is determined based on the axial positioning results, thereby adjusting the position of the mixing zone so that the intersection of the two ultrasonic guided wave signals coincides with the center of the layered damage in the axial direction, and multiple receiving points are arranged circumferentially along the fiber-wound tubular structure to receive the response signal.

[0093] In this embodiment, during circumferential positioning, a receiving point is arranged at one end of the fiber-wound tubular structure. A first set of time-domain signals is collected at the first receiving point, and a first dataset is obtained through the above preprocessing. A small angle increment ∆φ is added along the clockwise or counterclockwise direction, and a second set of time-domain signals is collected at the second receiving point. A second dataset is obtained through the above preprocessing.

[0094] Step S3-2: Input multiple datasets into the circumferential localization prediction model to obtain multiple predicted values ​​of the circumferential angles of multiple layered damages.

[0095] In this step, the circumferential localization prediction model is a trained one-dimensional convolutional neural network model (1D-CNN). The preprocessed dataset is input into the model, which simultaneously performs angle prediction and waveform reconstruction, and finally outputs the predicted value of the angle between the receiver point and the layered damage in the circumferential direction. This predicted value corresponds to the circumferential position of the layered damage, that is, the φ coordinate.

[0096] Figure 3 This is a schematic diagram of the structure of the one-dimensional convolutional neural network model in this embodiment.

[0097] like Figure 3As shown, the one-dimensional convolutional neural network model includes a dual-channel input module, a feature extraction module, an adaptive average pooling module, a multi-task module, and an output module. The feature extraction module consists of four cascaded convolutional units. Each convolutional unit contains a 1D convolutional layer with a kernel size of 7, a rectified linear unit, a batch normalization layer, and a max pooling layer. The number of channels gradually increases from 2 to 32, 64, 128, and 256 to extract deep temporal features. The extracted temporal features are compressed into a global feature vector by the adaptive average pooling module. The multi-task module has two independent branches: an angle prediction branch and a waveform reconstruction branch. The angle prediction branch outputs the predicted difference between the φ coordinate of the receiver point and the φ coordinate of the layered impairment through a fully connected network containing dropout layers. The waveform reconstruction branch reshapes the waveform using another fully connected network to capture mixing components with cumulative effects.

[0098] When training the circumferential positioning prediction model, in the multi-task learning process that includes angle prediction and waveform reconstruction tasks, the mean squared error (MSE) is used to construct the loss function, and the mean squared error L of angle prediction is calculated separately. angle With waveform reconstruction mean square error L wave :

[0099] (7)

[0100] In the formula, N represents the batch size, and in this embodiment, N=32, y i and These represent the true and predicted values ​​of the difference between the receiving point φ coordinate and the layered damage φ coordinate, respectively.

[0101] In multi-task learning, it is difficult to directly determine the mean square error L of angle prediction. angle With waveform reconstruction mean square error L wave The weights are determined by the homoscedastic uncertainty. Therefore, this embodiment introduces homoscedastic uncertainty and constructs the following loss function:

[0102] (8)

[0103] In the formula, σ i L is a learnable parameter, initially set to 1; angle For the mean square error of angle prediction, Lwave The mean square error of waveform reconstruction has equal weights; lnσ i As a regularization term, it prevents the total loss function from becoming infinitely large or infinitely small.

[0104] Due to σ 2 i During the descent, the variable may enter the negative range and cause the program to crash. To solve this problem, a new variable is created, let s i =ln(σ 2 i After the above transformation, the final total loss function is:

[0105] (9)

[0106] In the formula, s1 and s2 are the learnable parameters for the angle prediction task and the waveform reconstruction task, respectively, and are homoscedastic uncertainty parameters, both with an initial value of 0. The loss function includes a regularization term. It is always a positive number, which can prevent the occurrence of loss functions that are infinitely large or infinitely small.

[0107] During model training, five-fold cross-validation and the Adam optimizer are employed. The dataset is divided into five subsets for model training and validation. The Adam optimizer iteratively updates the convolutional layer weights and parameters simultaneously. Through backpropagation and multiple iterations, a feedback adjustment mechanism is established to optimize the L... angle L wave Optimize s1 and s2. Use the root mean square error of the validation set as the metric to train the model with the best performance.

[0108] (10)

[0109] In this embodiment, the training of the one-dimensional convolutional neural network model (1D-CNN) is as follows: Multiple receiving points are arranged circumferentially on the surface of the fiber-wound tubular structure at 1° intervals. An arbitrary function generator sequentially sets the phases of the fundamental waves a and b to the four phase combinations described above. The time-domain signals U of the displacement components under the four phase combinations are collected along the r-axis, φ-axis, and z-axis through these receiving points. r U φ and U z These time-domain signals are preprocessed to obtain multiple datasets.

[0110] When two fundamental waves converge in a region containing delamination damage, they may produce difference-frequency or sum-frequency components with cumulative effects. If the ultrasonic guided wave mixing satisfies the matching criterion... (where k) a With k b Let k be the wavenumber of fundamental wave a and fundamental wave b, respectively. a±b(Given the wavenumbers of the generated sum-frequency or difference-frequency components), it is possible to generate difference-frequency or sum-frequency components with cumulative effects. Therefore, in the preprocessing, for the waveform reconstruction task, a matching criterion for ultrasonic guided wave counter-mixing is introduced, and difference-frequency or sum-frequency components with cumulative effects are selected with the help of a specific time window, thereby improving the accuracy of layered damage angle prediction.

[0111] Specifically, the dataset contains dual-channel data and corresponding labels.

[0112] For the tags, the φ coordinate of the layered damage (i.e., φ=0°) is used as the reference point to construct the tag. The tag range is from 0° to 180°. If the angle between two receiving points and the reference point is the same, the same tag is used for the time domain signals of these two receiving points. For example, two receiving points with φ coordinates of 1° and 359° respectively use the same tag 1°.

[0113] For dual-channel data, the generated difference frequency component or sum frequency component pattern is determined based on the wavenumbers of fundamental frequency a and fundamental frequency b. The corresponding group velocity is determined based on the difference frequency component or sum frequency component. Then, the size of the time window is set based on the group velocity. The time window is used to completely select the time-domain signal containing the difference frequency component or sum frequency component that is sensitive to delamination damage in the time-domain signal. In this embodiment, starting from 50μs, a time window of 200μs and a step size of 100μs are selected to select the time-domain signal. For the selected time-domain signal, according to the above formula (4), the phase reversal method is used to process the time-domain signal. The amplitude of all frequencies near the difference frequency and sum frequency is extracted within a specific bandwidth, and the amplitude is normalized so that the amplitude value ranges between -1 and 1, thereby obtaining multiple data points. Then, for two receiving points with the same angle as the reference point, their two columns of data points are combined into dual-channel data, and the same label is used, with the angle of one of the receiving points as the label.

[0114] In this embodiment, the first dataset and the second dataset are respectively input into the trained circumferential positioning prediction model, and the model outputs the first predicted value respectively. Second predicted value .

[0115] Step S3-3: The angle increment discrimination method is used to determine the true circumferential angle of the layered damage based on multiple predicted values, thereby determining the circumferential position of the layered damage.

[0116] As mentioned above, the predicted values ​​output by the model correspond to the φ coordinates of the layered damage. However, the predicted value for one angle corresponds to two φ coordinates of the layered damage. Therefore, the angle increment discrimination method is further used to determine the true φ coordinates of the layered damage.

[0117] (11)

[0118] In the formula, The circumferential angle between the second receiving point and the delamination damage. The first predicted value, The second predicted value is ∆φ, which is a predetermined small angle increment.

[0119] By following the steps above, delamination damage caused by low-energy impacts in fiber-wound tubular structures can be detected and accurately located in the axial and circumferential directions.

[0120] In this embodiment, the above method is verified through simulation.

[0121] In the simulation, the commercial finite element simulation software Abaqus / EXPLICIT was used to construct a fiber-wound steel pipe as a fiber-wound tubular structure. This pipe consists of an outer layer of glass fiber reinforced polymer (GFRP) pipe and an inner layer of stainless steel pipe, with the GFRP pipe and steel pipe nested together. Furthermore, to simulate delamination damage caused by low-energy impact, a damage region A was cut out on the inner surface of the GFRP pipe, and a damage region B was cut out on the outer surface of the steel pipe. In the simulation, the tangential and normal behaviors of the surfaces of damage regions A and B were set to frictionless and "hard" contact; the constraints between other areas of the inner surface of the GFRP pipe and other areas of the outer surface of the steel pipe were set to bonded.

[0122] For example, the length, outer diameter, and thickness of the GFRP pipe are 3000 mm, 23 mm, and 0.5 mm, respectively; the length, outer diameter, and thickness of the steel pipe are 3000 mm, 22 mm, and 0.5 mm, respectively; and the length and width of the damaged area A and the damaged area B are 10 mm and 10 mm, respectively.

[0123] The material properties of GFRP pipes and stainless steel pipes are shown in Table 1 below:

[0124] Table 1 Material property parameters of GFRP and steel

[0125] GFRP 1960 48.6 8.5 0.33 0.38 3.5 3.0 Steel 7860 199.6 0.29

[0126] The material properties shown in Table 1 include density ρ, longitudinal elastic modulus E1 / E, transverse elastic modulus E2 and E3, and Poisson's ratio v. 12 and v 13 Poisson's ratio v 23 In-plane shear modulus G 12 and G 13 In-plane shear modulus G 23 In this context, the subscripts represent directions: direction 1 is the circumferential direction of the pipe, direction 2 is the axial direction of the pipe, and direction 3 is the radial direction of the pipe.

[0127] In step S1-1, the aforementioned material performance parameters and dimensions of the glass fiber wound steel pipe are obtained, and a cylindrical coordinate system is constructed. Let the center of the left end face of the glass fiber wound steel pipe be the origin, the r-axis (radial direction of the tubular structure) be inside the left end face, φ be the rotation angle of the r-axis, and the z-axis (axial direction of the tubular structure) point to the center of the right end face of the glass fiber wound steel pipe.

[0128] In step S1-2, based on the non-dispersion property of the T(0,1) mode, fundamental wave a and fundamental wave b select the T(0,1) mode for opposite propagation, and the location of the receiving point is arranged according to the mode.

[0129] Figure 4 This is a schematic diagram of the fiber-wound tubular structure with layered damage and the excitation signal in this embodiment. Figure 4 The middle section also shows the displacement field of the interaction between delamination damage and the opposing mixing of ultrasonic guided waves in a fiber-wound tubular structure.

[0130] like Figure 4 As shown, in steps S1-3, an amplitude of 1×10 is applied to the left and right end faces of the glass fiber wound steel pipe based on a cylindrical coordinate system. -7 A circumferential displacement *m* is used to excite the fundamental waves *a* and *b*. The excitation signal for fundamental wave *a* is a 10-cycle, Hanning-windowed sine wave with a center frequency of 100 kHz; the excitation signal for fundamental wave *b* is a 15-cycle, Hanning-windowed sine wave with a center frequency of 150 kHz. The maximum mesh size in the finite element simulation needs to be less than λ. min / 20, where λ min The wavelength is the shorter of the fundamental wavelengths a and b, set to 0.5 mm in the simulation. The simulation time step needs to be less than 1 / 20 / f. max , where f max The maximum center frequency among the center frequencies of fundamental frequency a and fundamental frequency b is used, and the time step in the simulation is set to 1×10⁻⁶. -8 s.

[0131] Receiver points are placed outside the mixing region, specifically one at each of the following locations: φ = 90°, r = 11.5mm, z = 800mm and φ = 90°, r = 11.5mm, z = 2200mm. Fundamental frequencies a and b mix at the delamination damage site, generating difference frequency components and sum frequency components that propagate simultaneously to the left and right ends of the fiberglass-wound steel pipe. The displacement field resulting from the interaction between the ultrasonic guided wave and the delamination damage is as follows: Figure 4 As shown in the image.

[0132] In step S2-1, the time delay t between fundamental frequency a and fundamental frequency b is changed. dThis changes the position of the mixing zone, enabling axial scanning of the entire fiberglass-wound steel tube. The time delay t varies depending on the length of the steel tube. d The interval is 50µs, ranging from 50µs to 200µs. The intersection position moves from z = 1220mm to z = 1780mm to achieve axial scanning of the entire steel pipe.

[0133] In step S2-2, four physical models were constructed. The four original time-domain signals were processed using the phase reversal method to eliminate the fundamental frequency and second harmonic, highlighting the difference frequency component or sum frequency component generated due to layering damage. Specifically, in the first and second physical models, the excitation signals of fundamental frequency a and fundamental frequency b have the same phase, equal to 0° and 180°, respectively; in the third physical model, the excitation signal of fundamental frequency a has a phase of 0°, and the excitation signal of fundamental frequency b has a phase of 180°; in the fourth physical model, the excitation signals of fundamental frequency a and fundamental frequency b have exactly opposite phases.

[0134] Figure 5 This is an example diagram of the original time-domain signal in this embodiment. Figure 6 This is an example diagram of the spectrum of the original time-domain signal in this embodiment. Figure 7 This is an example diagram of the time-domain signal after phase reversal processing in this embodiment. Figure 8 This is an example diagram of the spectrum of the time-domain signal after phase reversal processing in this embodiment.

[0135] like Figures 5 to 8 As shown, taking a typical time-domain signal at φ = 90°, r = 11.5mm, and z = 2200mm as an example, it can be seen that after phase reversal processing, the difference frequency component or sum frequency component that was originally submerged has become prominent.

[0136] In step S2-2, the normalized nonlinear acoustic energy β of the difference frequency component or sum frequency component is then calculated. energy For the original signal processed by the phase reversal method, a Fourier transform is performed. The components near the difference frequency and the components near the sum frequency (with a bandwidth of 20kHz) are extracted using the gate function G(f). The sum of the squares of the amplitudes of these components is calculated as β. energy .

[0137] Figure 9 This is a graph showing the normalized nonlinear acoustic energy of the intersection location and the difference frequency component in this embodiment. Figure 10 This is a graph showing the intersection location and the normalized nonlinear acoustic energy of the sum-frequency components in this embodiment.

[0138] like Figure 9 and Figure 10 As shown, in steps S2-3, the normalized nonlinear acoustic energy β is plotted. energyUsing the vertical axis as the vertical axis and the intersection position Z as the horizontal axis, m For the curve on the horizontal axis, plot the time-domain signal U of the displacement component for both the difference frequency component and the sum frequency component. r U φ and U z The corresponding curve.

[0139] from Figure 9 and Figure 10 As can be seen, when the mixing region coincides with the location of the delamination damage, difference frequency components and sum frequency components appear. The closer the intersection location is to the center of the delamination damage, the higher the normalized nonlinear acoustic energy β of the difference frequency components and sum frequency components. energy The larger it is. In the diagram, U... φ The corresponding β energy The peak value is reached at z = 1500mm. Therefore, in step S2-4, the z coordinate of the delamination damage in the glass fiber wound steel pipe can be determined by the curve, thereby realizing the axial positioning of the delamination damage.

[0140] Additionally, in this example, U based on the sum-frequency component z The corresponding β energy The z-coordinate of delamination damage in glass fiber wound steel pipes can be roughly determined, while the U-coordinate based on the difference frequency component can be used to determine the z-coordinate of the delamination damage. r The corresponding β energy U z The corresponding β energy and the sum-frequency component U r The corresponding β energy Therefore, it is impossible to determine the z-coordinate of the delamination damage in the glass fiber wound steel pipe.

[0141] In step S3-1, as described above, a first dataset and a second dataset are constructed for the two receiving points respectively.

[0142] In step S3-2, the first and second datasets are input into the trained one-dimensional convolutional neural network model, respectively, and one column of their amplitudes is copied to form dual-channel data. The dual-channel data passes through the feature extraction module to extract deep temporal features. The extracted features are compressed into a global feature vector by adaptive average pooling. Then, the angle prediction branch outputs the predicted value of the difference between the φ coordinate of the measurement point and the φ coordinate of the hierarchical damage through a fully connected network containing Dropout. For the first and second datasets, the first predicted value is output respectively. Second predicted value .

[0143] In this step, the training of the circumferential positioning prediction model is as follows. Along the circumference of the glass fiber wound steel pipe, starting from φ = 90°, r = 11.5mm, z = 2200mm, the time-domain signal U of the displacement component is received every 1°.r U φ and U z Then, the data is preprocessed by combining the two amplitude columns from two receivers with the same angle to the reference point into dual-channel data, using the same label. For example, the two amplitude columns from two receivers with φ coordinates of 1° and 359° are combined into dual-channel data with the same label indicating an angle of 1°, thus forming a dataset that serves as input to the circumferential positioning prediction model. Specifically, for two receivers with φ coordinates of 0° and 180°, one of their own amplitude columns is copied to form dual-channel data.

[0144] Furthermore, for the original signal processed by the phase reversal method, bandpass filtering is performed and normalization is applied about the maximum value, with all amplitudes between -1 and 1. This embodiment starts from 50μs, using a 200μs time window and a 100μs step size to analyze the average recognition rate of the time-domain signal at different time intervals. The 200μs time-domain signal contains 425 data points. Interpolation is used to convert these 425 data points into 400 data points.

[0145] During model training, the training dataset is input into a one-dimensional convolutional neural network (1D-CNN) model. Leveraging its multi-task learning capabilities, the same 1D-CNN model simultaneously performs angle prediction and waveform reconstruction tasks for layered damage. Labeled dual-channel data is processed by a feature extraction module, with the number of channels gradually increasing from 2 to 32, 64, 128, and 256 to extract deep temporal features. The extracted features are then compressed into a global feature vector using adaptive average pooling, and subsequently branched into two independent branches. The angle prediction branch outputs the predicted difference between the φ coordinate of the receiver point and the φ coordinate of the layered damage through a fully connected network incorporating Dropout. The waveform reconstruction branch reshapes the waveform through another fully connected network to capture mixing components with cumulative effects, thus better optimizing model parameters during training. The model parameters are then updated using a loss function until the optimal-performing model is obtained, and the model parameters are saved.

[0146] Figure 11 This is a root mean square error (RMSE) plot of the angle prediction of the time-domain signal in different time windows in this embodiment, where the horizontal axis represents different time windows and the vertical axis represents the RMSE value.

[0147] like Figure 11 As shown, for time-domain signals in the 900μs-1100μs range, the RMSE is 2.81°, which is better than that for time-domain signals in other time windows. This is because in this example, the phase-matching criterion is satisfied and the arrival time of the frequency components is between 900μs and 1100μs.

[0148] Figure 12This is a fitting graph of the actual angle and the predicted angle of the layered damage in this embodiment.

[0149] like Figure 11 and Figure 12 As shown, on the validation set, for true angles ranging from 0° to 180°, the predicted angles generally fall within the error band, with very small deviations from the true angles. Therefore, simulations have verified that the method described in this embodiment is effective for time-domain signals of 900μs-1100μs, with a simultaneous input time-domain signal U. r U φ and U z It accurately predicted the difference between the φ coordinate of the receiving point and the φ coordinate of delamination damage in the glass fiber wound steel pipe.

[0150] Furthermore, in addition to the angle prediction method in the embodiments, the inventors have designed three other angle prediction methods for comparison. The angle prediction method in the embodiments, as described above, uses the time-domain signal processed by the phase inversion method as the input signal and employs a one-dimensional convolutional neural network model with multi-task learning (including angle prediction and waveform reconstruction tasks). The method in the embodiments is denoted as Pulse-inversion+Multi-task. The three other angle prediction methods for comparison include:

[0151] Method 1 uses an original time-domain signal as the input signal and employs a 1D-CNN that only includes the angle prediction task. This method is referred to as Origin signals.

[0152] Method 2 uses four original time-domain signals with different initial phases as input signals and employs a 1D-CNN that only includes the angle prediction task. This method is referred to as Four cases.

[0153] Method 3 uses the time-domain signal processed by the phase inversion method as the input signal and employs a 1D-CNN that only includes the angle prediction task. This method is denoted as Pulse-inversion.

[0154] The four methods described above were used to predict angles on the validation set, and the performance of the four methods was evaluated using the following four metrics: root mean square error (RMSE), mean absolute error (MAE), maximum error (MaxError), and coefficient of determination (R²). 2 The calculation formula is as follows:

[0155] (12)

[0156] In the formula, N is the total number of samples, and y i , and These represent the true value, predicted value, and arithmetic mean of the difference between the φ coordinate of the receiving point and the φ coordinate of the layered damage, respectively.

[0157] Figure 13 This is a comparison chart of four indicators for the four angle prediction methods in this embodiment.

[0158] like Figure 13 As shown, the comparison results indicate that the average recognition rate of Method 3 (Pulse-inversion) is higher than that of Method 1 (Origin signals). Compared with Method 1, the root mean square error (RMSE) of Method 3 decreases from 7.35° to 6.34°. The four metrics of the method (Pulse-inversion + Multi-task) in this embodiment all optimize the corresponding metrics of the other three methods. Finally, in step S3-3, the true φ coordinates of the layered damage are determined using the angle increment discrimination method. For the first predicted value of the first receiving point... The second predicted value of the second receiving point Increase the angle increment ∆φ in the clockwise direction by a small amount, and use the formula The circumferential angle between the second receiving point and the delamination damage was obtained. Ultimately, this achieves circumferential localization of layered damage.

[0159] As described above, in this embodiment, by adjusting the time delay between fundamental wave a and fundamental wave b, and utilizing the multi-task learning capability of the one-dimensional convolutional neural network model and the angle increment discrimination method, the axial and circumferential positioning of delamination damage caused by low-energy impact in glass fiber wound steel pipes was successfully achieved.

[0160] The role and effect of the embodiments

[0161] The ultrasonic guided wave mixing circumferential localization method for delamination damage in fiber-wound tubular structures provided in this embodiment utilizes mixing excitation to form a mixing region within the fiber-wound tubular structure. For delamination damage caused by low-energy impacts that cannot be detected or accurately located using conventional methods, the method leverages the local contact acoustic nonlinearity to generate mixing components when the mixing region coincides with the delamination damage. Therefore, it can rely on these mixing components to achieve accurate detection and localization of such delamination damage, including axial and circumferential localization. Furthermore, by employing a deep learning model to predict the circumferential angle of the delamination damage and combining it with an angle increment discrimination method to determine the true circumferential angle, it achieves end-to-end circumferential angle prediction of the delamination damage, avoiding human error and improving the efficiency of non-destructive testing and localization.

[0162] In this embodiment, two torsional modes with different center frequencies and opposite directions are used. The spatial intersection position of the two ultrasonic guided waves can be changed by using only the time delay parameter, thereby enabling continuous axial scanning along the entire length with high axial positioning accuracy. Furthermore, by setting the center frequency, the problem of high-order harmonic interference can be avoided.

[0163] Furthermore, for the axial localization of delamination damage, normalized nonlinear acoustic energy is introduced. Compared with traditional acoustic nonlinear parameters, it can suppress the effects of noise and frequency shift. At the same time, phase reversal technology is used to retain only the delamination damage-related and mixing components, which greatly filters out interference and improves the signal-to-noise ratio, thus improving the robustness of detection and localization.

[0164] Furthermore, the circumferential positioning prediction model employs a multi-task one-dimensional convolutional neural network model. During model training, angle prediction and waveform reconstruction are performed simultaneously. Waveform reconstruction assists in constraining the convolutional network, enabling it to focus on extracting features from mixing components that satisfy the matching criterion and filtering out invalid noise components, thus outputting more accurate circumferential angle predictions. In addition, the introduction of a homoscedasticity uncertainty parameter allows the model to adaptively optimize the dual-task loss weights, saving time on repeated manual debugging and enabling more efficient model training.

[0165] The above embodiments are merely illustrative of specific implementations of the present invention, and the present invention is not limited to the scope of the description of the above embodiments. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are only for illustrating the principles of the present invention. Various changes and modifications can be made to the present invention without departing from the spirit and scope thereof, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

[0166] For example, in the above embodiments, this method is used to detect and locate delamination damage in glass fiber wound steel pipes. In fact, this method can also be used to detect and locate delamination damage in other types of fiber wound tubular structures.

Claims

1. A method of ultrasonic guided wave circumferential positioning of delamination damage in fiber-wound tubular structures, the method comprising: Includes the following steps: Step S1: Obtain the material and geometric parameters of the fiber-wound tubular structure, and apply a mixing excitation to the fiber-wound tubular structure using a nonlinear ultrasonic measurement system based on these parameters to form a mixing region in the fiber-wound tubular structure. Step S2: Adjust the position of the mixing region in the axial direction of the fiber-wound tubular structure by changing the parameters of the mixing excitation, and locate the delamination damage axially based on the response signal after mixing. Step S3: Based on the axial positioning result, the mixing region is placed at the location of the delamination damage, and the delamination damage is circumferentially located based on the mixed response signal. In step S3, the circumferential positioning prediction model is used to obtain multiple predicted values ​​of the circumferential angle of the layered damage based on the response signal, and the angle increment discrimination method is used to distinguish the true circumferential angle among the multiple predicted values, thereby realizing the circumferential positioning.

2. The ultrasonic guided wave mixing circumferential localization method for layered damage of fiber-wound tubular structures according to claim 1, characterized in that: wherein Step S1 includes the following sub-steps: Step S1-1: Obtain the material parameters and geometric parameters of the fiber-wound tubular structure; Step S1-2: Based on the material parameters and the geometric parameters, select the mode pair for the opposing mixing of two ultrasonic guided waves, and arrange the receiving point position based on the selected mode pair; Steps S1-3: Using the nonlinear ultrasonic measurement system, two ultrasonic guided wave signals with different center frequencies and opposite directions are applied to both ends of the fiber-wound tubular structure to form the mixing zone.

3. The ultrasonic guided wave mixing circumferential location method for delaminated damage of fiber-wound pipe-like structures of claim 2, Its features are: Step S2 includes the following sub-steps: Step S2-1: By changing the time delay between the two ultrasonic guided wave signals, the position of the mixing region is adjusted along the axial direction of the fiber-wound tubular structure, thereby achieving axial scanning. Step S2-2: During the axial scanning process, the response signals of multiple different receiving points along the axial direction are acquired, and the normalized nonlinear acoustic energy at each intersection position of the two ultrasonic guided waves is calculated based on the multiple response signals. Step S2-3: Generate the relationship curve between the intersection location and the normalized nonlinear acoustic energy; Step S2-4: Based on the relationship curve, the functional relationship between the layered damage and the normalized nonlinear acoustic energy, determine the axial position of the layered damage.

4. The ultrasonic guided wave mixing circumferential localization method for layered damage of fiber-wound tubular structures according to claim 3, characterized in that: in, In step S1-2, the non-dispersion torsional mode T(0,1) is selected. In steps S1-3, a center frequency of f is applied to one end of the fiber-wound tubular structure. a The number of cycles is n a The fundamental wave a, with a center frequency f applied at its other end b The number of cycles is n b The fundamental wave b, where both fundamental wave a and fundamental wave b are ultrasonic guided waves, are received at both ends of the fiber-wound tubular structure. In step S2-1, a time delay t is provided between the fundamental a and the fundamental b d The intersection position is expressed as: In the formula, l is the length of the fiber-wound tubular structure. , Let be the group velocities of fundamental wave a and fundamental wave b, respectively. In step S2-2, the original time-domain signal is extracted from the response signal and subjected to phase inversion processing. A Fourier transform is performed on the phase-inverted original time-domain signal to obtain its spectrum. A gate function is used to extract the amplitudes of the difference frequency and all frequencies near the sum frequency within a specific bandwidth of this spectrum. The integral of the squared amplitudes is then calculated as the normalized nonlinear acoustic energy. In the formula, A(f) is the spectrum of the time-domain signal after phase reversal processing, G(f) is the gate function, and f low and f high These are the lower and upper frequency limits for a specific bandwidth. In steps S2-4, based on the relationship curve, the functional relationship between the layered damage and the normalized nonlinear acoustic energy, the z-coordinate of the layered damage is obtained, wherein the z-axis is consistent with the axial direction of the fiber-wound tubular structure.

5. The ultrasonic guided wave mixing circumferential localization method for layered damage of fiber-wound tubular structures according to claim 4, characterized in that: wherein In steps S1-3, the difference frequency and sum frequency of the center frequencies of the fundamental frequency a and the fundamental frequency b avoid the center frequency f. a Or the center frequency f b frequency multiplication, In step S2-1, the time delay is applied only to one of the fundamental frequency a and the fundamental frequency b, and in a specific step size. An axial scan is performed on the fiber-wound tubular structure, wherein c g Let Δt be the group velocity of the fundamental wave a and the fundamental wave b. d This represents the change in time delay.

6. The ultrasonic guided wave mixing circumferential localization method for layered damage of fiber-wound tubular structures according to claim 4, characterized in that: wherein In step S2-2, the phases of the fundamental wave a and the fundamental wave b are set to 0° and 0°, 0° and 180°, 180° and 0°, and 180° and 180°, respectively, and the corresponding original time-domain signals S are acquired. a,b (t), S a,-b (t), S -a,b (t) and S -a,-b (t), The original time-domain signal after phase reversal processing is represented as follows: 。 7. The ultrasonic guided wave circumferential location of delaminated damage in fiber-wound pipe like structures method of claim 3, Its features are: Step S3 includes the following sub-steps: Step S3-1: Place the mixing region at the axial position of the layered damage. For each receiving point, collect the response signals of multiple different receiving points in the circumferential direction, and construct multiple datasets based on the multiple sets of response signals. Step S3-2: Input multiple datasets into the circumferential localization prediction model to obtain multiple predicted values ​​of the circumferential angle of the layered damage, wherein the circumferential localization prediction model is a trained one-dimensional convolutional neural network model. Step S3-3: Using the angle increment discrimination method, the true circumferential angle of the layered damage is determined based on multiple predicted values, thereby determining the circumferential position of the layered damage.

8. The ultrasonic guided wave mixing circumferential localization method for layered damage of fiber-wound tubular structures according to claim 7, characterized in that: wherein In step S3-1, the time delay is determined based on the axial positioning result, thereby adjusting the position of the mixing region so that the intersection position coincides with the center of the layered damage in the axial direction. Two sets of time-domain signals are acquired at the first receiving point and the second receiving point, respectively. After preprocessing the two sets of time-domain signals, the corresponding first dataset and second dataset are obtained. In step S3-2, the first dataset and the second dataset are respectively input into the circumferential positioning prediction model to obtain the first and second predicted values ​​of the circumferential angle. In step S3-3, the true φ coordinates of the layered damage are determined based on the first predicted value and the second predicted value using the angle increment discrimination method: In the formula, The circumferential angle between the second receiving point and the delamination damage. The first predicted value, The second predicted value is ∆φ, where ∆φ is a predetermined small angular increment, φ is the rotation angle of the r-axis, and the r-axis is aligned with the radial direction of the fiber-wound tubular structure.

9. The ultrasonic guided wave mixing circumferential localization method for delamination damage of fiber-wound tubular structures according to claim 8, characterized in that: wherein, In step S3-1, the constructed dataset contains dual-channel data. In step S3-2, the one-dimensional convolutional neural network model has a dual-channel input module, a feature extraction module, an adaptive average pooling module, and a multi-task module. The multi-task module includes an angle prediction branch and a waveform reconstruction branch. During model training, the dataset is input into the one-dimensional convolutional neural network model. The predicted value is output through the angle prediction branch, and the waveform is reshaped through the waveform reconstruction branch to capture the difference-frequency component or sum-frequency component with cumulative effect. Five-fold cross-validation is used, and the weights and parameters of the convolutional layer are iteratively updated based on the loss function by the Adam optimizer. The loss function is expressed as: In the formula, L angle For the mean square error of angle prediction, L wave The mean square error for waveform reconstruction. S1 and S2 are the regularization terms, and are the learnable parameters for the angle prediction task and waveform reconstruction task, respectively, and are homoscedastic uncertainty parameters.

10. The ultrasonic guided wave mixing circumferential localization method for delamination damage of fiber-wound tubular structures according to claim 9, characterized in that: wherein For the dual-channel data, the mode of the generated mixing component is determined by using the phase matching criterion based on the wavenumbers of the fundamental frequency a and the fundamental frequency b. The corresponding group velocity is determined based on the mixing component. The size of the time window is then set based on the group velocity. The time window is used to select the time-domain signal containing the mixing component that is sensitive to the layering damage in the original time-domain signal. After processing the time-domain signal using the phase inversion method, the amplitude of all frequencies near the difference frequency or sum frequency is extracted within a specific bandwidth. The amplitude is then normalized to obtain multiple data points.

Images