A dual-constraint aeroengine blade defect imaging method

Abstract

The application provides a double-constraint aero-engine blade defect imaging method, steps of which are as follows: three-dimensional modeling is performed on an aero-engine blade, a dispersion curve of the aero-engine blade is calculated, and an optimal frequency-mode combination corresponding to an excitation signal is screened out; the obtained optimal frequency-mode combination is set as a parameter of the excitation signal; a solid mechanics module is used as a physical field, an excitation position and a receiving position are set on the blade, simulation is performed to obtain an ultrasonic guided wave detection signal in the aero-engine blade; multiple simulations are performed to obtain a sound field distribution data set; a double-constraint aero-engine blade defect imaging network is constructed, the ultrasonic guided wave detection signal in the data set is pretreated, and the defect imaging network is trained; the real-time collected ultrasonic guided wave detection signal is input into the trained defect imaging network to obtain defect information. The aero-engine blade defect imaging is realized through the dispersion curve and the design of the double-constraint aero-engine blade defect imaging network.

Description

Technical Field

[0001] This invention relates to the technical field of ultrasonic imaging, and more particularly to a method for imaging defects in dual-constraint aero-engine blades. Background Technology

[0002] As the core power unit of an aircraft, the reliability of an aero-engine directly affects the operational safety of the aircraft. In the field of civil aviation, engine failure can not only lead to flight disruptions and economic losses amounting to tens of millions of yuan, but may even cause major accidents resulting in casualties. In complex service environments, blade surfaces must also cope with multiple damage mechanisms, including high-speed impacts from sand particles, foreign object impacts, and thermo-coupling effects. These extreme conditions can easily lead to the peeling and detachment of the thermal barrier coating on the blade surface, and induce progressive defects such as the initiation of internal microcracks or creep deformation, which may ultimately lead to catastrophic accidents such as blade fracture failure, engine surge, or even in-flight disintegration.

[0003] Aero-engine blade inspection technologies include various methods such as magnetic particle inspection, penetrant testing, eddy current testing, machine vision, radiographic testing, and ultrasonic testing. However, most of these blade defect detection methods based on traditional non-destructive testing techniques are limited to surface or near-surface defect detection. Further research is needed on the defect detection speed, safety, and quantitative detection of defects in some methods.

[0004] Ultrasonic guided wave testing technology, a branch of ultrasonic testing, overcomes the limitations of traditional ultrasonic testing, which requires movement along the structure to achieve large-area scanning. Through single-point excitation, mechanical waves undergo multiple reflections within the waveguide structure, generating superposition interference and geometric dispersion to form ultrasonic guided waves, thus enabling large-area scanning. Ultrasonic guided wave testing technology allows for rapid detection of waveguide structures, combining the safety and reliability advantages of traditional ultrasonic volume wave testing. It is widely used for defect detection in plates, pipes, and other complex structures. To more intuitively and accurately display and assess the damage to the structure under test, the industry has developed several imaging algorithms, such as full waveform inversion, robust breast ultrasound tomography hybrid algorithms (extending from medicine to the non-destructive testing field), damage probability detection algorithms, and reverse time migration methods. These methods offer the advantage of quantitative defect detection and imaging compared to traditional non-destructive testing techniques. While ultrasonic guided wave testing technology, such as full waveform inversion, solves the problems of traditional non-destructive testing techniques, a major limitation is that these methods typically only image defects in plate or tubular structures. Summary of the Invention

[0005] To address the technical problem that existing non-destructive testing methods for aero-engine blades are mostly limited to surface or near-surface defect detection, this invention proposes a dual-constraint aero-engine blade defect imaging method. By employing the semi-analytical finite element method to obtain the optimal frequency-modal combination of the blade under a three-dimensional curvilinear coordinate system, and designing a dual-constraint aero-engine blade defect imaging network, defect imaging of aero-engine blades can be achieved.

[0006] To achieve the above objectives, the technical solution of the present invention is as follows: a dual-constraint aero-engine blade defect imaging method, the steps of which are as follows:

[0007] Step 1: Perform 3D modeling of the aero-engine blade, calculate the ultrasonic guided wave dispersion curve of the aero-engine blade, and select the optimal frequency-mode combination corresponding to the excitation signal.

[0008] Step 2: Set the obtained optimal frequency-mode combination as the parameters of the excitation signal; use the solid mechanics module as the physical field, set the excitation and receiving positions on the blade, and perform simulation to obtain the ultrasonic guided wave detection signal in the aero-engine blade; perform multiple simulations to obtain the sound field distribution dataset;

[0009] Step 3: Construct a dual-constraint aero-engine blade defect imaging network. Preprocess the ultrasonic guided wave detection signals in the dataset and train the defect imaging network. Input the real-time acquired ultrasonic guided wave detection signals into the trained defect imaging network to obtain defect information and realize defect imaging of aero-engine blades.

[0010] Preferably, the method for three-dimensional modeling of the aero-engine blade is as follows: performing a three-dimensional scan of the aero-engine blade to obtain the point cloud coordinates of the blade; performing simulation on three-dimensional finite element software, and using the point cloud coordinates to perform three-dimensional modeling of the aero-engine blade to obtain a three-dimensional curve.

[0011] Preferably, the coordinates of the three-dimensional curvilinear coordinate system are... Where x and y represent cross sections perpendicular to the blade axis, and s represents the blade axis; the position vector in the three-dimensional curve. Through Descartes orthogonal bases Represented as:

[0012] ;

[0013] Where r and d are the radius and helical step of the curve in the Cartesian coordinate system, and l is the length of the curve in the helical step.

[0014] When the radius r = 0, the spiral centerline is the torsion centerline, and the three-dimensional curvilinear coordinate system is used... This indicates that for any position vector in the Cartesian coordinate system... Represented as:

[0015] .

[0016] Preferably, the dispersion curve of the aero-engine blade is solved using the semi-analytical finite element method.

[0017] Preferably, in the semi-analytical finite element method, the cross-sectional domain or line domain is discretized and divided into a planar mesh; when the waveguide cross-section is symmetrical, the semi-analytical finite element formula for the one-dimensional mesh is the governing equation of the ultrasonic guided wave: ;in, For semi-analytical finite element mesh The nodal displacements; the matrix and Determined by the waveguide cross-section; Indicates wave number, Representation matrix The dimension is , and All are stiffness matrices. A symmetric matrix representing undamped motion;

[0018] For each angular frequency The wave number k is obtained from the control equation of the ultrasonic guided wave, and thus the phase velocity is obtained:

[0019] ;

[0020] angular frequency With phase velocity c p The relationship is represented by a dispersion curve.

[0021] Preferably, based on the convergence criterion of shear wave velocity in elastic materials, the maximum mesh length of the semi-analytical finite element method is: ;in, Maximum input frequency, It is a constant. Indicates the shear wave velocity;

[0022] Since different angular frequencies correspond to different numbers of modes, the angular frequency corresponding to the lowest number of modes is selected as the optimal angular frequency for the excitation signal. The mode corresponding to the optimal angular frequency is the mode with the largest slope in the dispersion curve. The optimal frequency-mode combination is within 20000 Hz - 2 × 10⁻⁶. 6 Dispersion curves in the Hz range.

[0023] Preferably, a three-dimensional finite element model is used to simulate an aero-engine blade model with defects such as holes, creep, scratches, tears, and fatigue cracks, and material properties are set; boundary conditions of the wave equation are defined, and the perfect matching layer technology is used to suppress the boundary reflection of the blade, and the blade model is meshed; the frequency and mode are directly obtained as parameters of the excitation signal; after setting the parameters of the excitation signal according to the optimal frequency-mode combination determined in the dispersion curve, the wave equation is solved to obtain the ultrasonic guided wave sound field distribution in the aero-engine blade, and the sound field distribution is the ultrasonic guided wave detection signal;

[0024] Multiple simulations were performed on defective aero-engine blades to obtain simulation datasets. Experiments were conducted according to the simulation settings to establish experimental datasets. Simulation and experimental datasets were used to verify each other.

[0025] Preferably, the preprocessing performs feature dimensionality reduction on the hybrid dataset obtained from the simulation dataset and the experimental dataset to obtain a standardized signal dataset; based on the standardized signal dataset, a dual-constraint aero-engine imaging network is designed, including a multi-scale residual convolution module; and a composite loss function is used to train the dual-constraint aero-engine imaging network.

[0026] Preferably, the feature dimensionality reduction processing method is as follows: extract the first arrival wave of the ultrasonic guided wave detection signal and remove redundant signal segments; perform a fast Fourier transform on the first arrival wave to obtain the amplitude spectrum and phase spectrum; extract the feature amplitude and phase information corresponding to the excitation center frequency from the amplitude spectrum and phase spectrum, and construct a three-dimensional feature matrix, wherein the first and second dimensions represent the feature amplitude and phase, and the third dimension represents different modes; use a zero-mean normalization algorithm to standardize the three-dimensional feature matrix to obtain a standardized signal dataset;

[0027] The multi-scale residual convolution module includes a multi-scale branch structure, which contains N parallel convolution branches. Each convolution branch uses convolution kernels of different scales to extract multi-scale features; and performs multi-scale residual convolution on the multi-scale features to obtain defect information.

[0028] Preferably, the excitation position and the receiving position are set around the entire blade at a distance of 5 mm from the blade edge;

[0029] The mesh is divided into quadrilateral discrete or triangular discrete;

[0030] The method for mutual verification between simulation and experiment is as follows: calculate the correlation coefficient between the simulation dataset and the experimental dataset. If the correlation coefficient is greater than 0.75, the data obtained from the simulation and experiment are considered to be correct. Verify whether the frequency-modal combination for defect detection is optimal. If it is not optimal, select the angular frequency corresponding to the condition with fewer modes as the optimal frequency of the excitation signal and select the mode corresponding to the optimal frequency.

[0031] The multi-scale features Where X is a three-dimensional feature matrix, X i For multi-scale features, Let k represent the convolution operation in the i-th branch. i The kernel size is usually an odd number, such as 3, 5, or 7, which are commonly used.

[0032] Defect information is obtained from the output using scale residual convolution. Where Attention represents the attention mechanism, and Concat represents the concatenation operation along the channel dimension. For Hadamah accumulation;

[0033] The composite loss function is: ;

[0034] Wherein, loss L represents the error between the true label and the predicted defect, loss L sparse α represents the regularization term for L1 sparseness, and α represents the regularization strength.

[0035] By using the adaptive moment estimation algorithm to minimize the composite loss function, when the composite loss function is minimized, the error between the true label and the predicted defect information and the L1 sparse regularization term are both small. It is assumed that the true label and the predicted defect information are close, and the defect is accurately predicted, thus realizing defect imaging of aero-engine blades.

[0036] Compared with existing technologies, the beneficial effects of this invention include: calculating the ultrasonic guided wave dispersion curve of aero-engine blades; solving the wave equation after setting the excitation signal parameters based on the optimal frequency-mode combination determined in the dispersion curve to obtain the ultrasonic guided wave sound field distribution in the aero-engine blade; preprocessing the ultrasonic guided wave detection signal; constructing a dual-constraint aero-engine blade defect imaging network and training the network; establishing a nonlinear mapping relationship from signal characteristics to aero-engine blade defects to achieve defect imaging of aero-engine blades. This invention, based on ultrasonic guided wave detection, is no longer limited to defect imaging of plate-shaped or tubular structures; it can detect defects in curved structures and achieve defect imaging of aero-engine blades. Furthermore, this invention considers the curved surface of the blade when solving the dispersion curve using a semi-analytical finite element method, which is more accurate than approximating the blade as a flat plate to calculate the dispersion curve. Attached Figure Description

[0037] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0038] Figure 1 This is a flowchart of the present invention.

[0039] Figure 2 This describes the calculation process of the ultrasonic guided wave dispersion curve and sound field solution for the aero-engine blade in this invention.

[0040] Figure 3 This is the imaging process for defects in aero-engine blades in this invention.

[0041] Figure 4 This refers to the detection signal for the aero-engine blades in this invention. Detailed Implementation

[0042] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0043] like Figure 1 As shown, a dual-constraint aero-engine blade defect imaging method includes the following steps:

[0044] Step 1: Perform 3D modeling of the aero-engine blade to obtain the dispersion curve, calculate the ultrasonic guided wave dispersion curve of the aero-engine blade, and select the optimal frequency-mode combination corresponding to the excitation signal.

[0045] a. In order to accurately solve the dispersion curve of aero-engine blades, a three-dimensional model of the aero-engine blades is performed.

[0046] A 3D scan of the aero-engine blade is performed to obtain its 3D point cloud coordinates. Simulation is then conducted using the 3D finite element software COMSOL Multiphysics, where the aero-engine blade is modeled in 3D using these point cloud coordinates. This step is crucial for subsequent calculation of the dispersion curve, from which the excitation frequency and modes can be obtained. Knowing the excitation frequency and modes allows for their use in the simulation, ultimately leading to the acquisition of the ultrasonic guided wave detection signal.

[0047] Let the coordinates of the three-dimensional curvilinear coordinate system be... Where x and y represent cross sections perpendicular to the blade axis, and s represents the blade axis; the position vector in the three-dimensional curve can be obtained through a Cartesian orthogonal basis. express:

[0048] (1)

[0049] Where r and d are the radius and helical step of the curve in the Cartesian coordinate system, and l is the curve length of the helical step. This represents a position vector.

[0050] When r=0, the spiral centerline is the torsion centerline, thus allowing the three-dimensional curvilinear coordinate system to be used... This indicates that for any position vector in the Cartesian coordinate system... It can be represented as:

[0051] (2)

[0052] The obtained position vectors are used as modeling data in the semi-analytical finite element solution modeling.

[0053] b. Based on the three-dimensional curve of the modeled aero-engine blade, solve the dispersion curve of the aero-engine blade using semi-analytical finite element method.

[0054] In the semi-analytical finite element method, the cross-sectional or linear domain is discretized and divided into a planar mesh. When the waveguide cross-section is symmetrical, a semi-analytical finite element formula specific to a one-dimensional mesh can be used. Typically, the governing equations for ultrasonic guided waves are:

[0055] (3)

[0056] in, For semi-analytical finite element mesh The nodal displacements; the matrix and It is determined by factors such as the size, shape, isotropic or anisotropic properties, and angular frequency of the waveguide cross-section; Indicates wave number, Representation matrix The dimension is , and These are all stiffness matrices, and are all related to the y, s, and x directions. A symmetric matrix representing undamped motion.

[0057] For each angular frequency According to equation (3), the corresponding wave number k can be obtained, and thus the phase velocity c can be obtained. p :

[0058] (4)

[0059] angular frequency With phase velocity c p The relationship is represented by the dispersion curve.

[0060] In semi-analytical finite element method (SEM), the number of meshes significantly affects the accuracy of the solution; as the number of meshes increases, the computational cost also increases; therefore, it is necessary to consider both the accuracy of the solution and the computational cost. For the generation of an exact solution, the number of meshes in a semi-analytical finite element method is related to the ultrasonic guided wave at the maximum frequency; as the frequency increases, each wave has more displacement fields with crests and troughs on its cross-section, which leads to increased solution errors at high frequencies; the wave field is also related to the wave type. For a given frequency, a specific mesh is more suitable for certain modes, thus resulting in higher solution accuracy in these modes. Based on the convergence criterion of shear wave velocity in elastic materials, the maximum mesh length of a semi-analytical finite element method can be calculated using the following formula:

[0061] (4)

[0062] in, Maximum input frequency, The maximum grid length is a constant. It is usually 4 or 10. The shear wave velocity is determined by the shear modulus and density of the material; for viscoelastic materials, since the shear wave wavelength remains almost constant, this standard still applies.

[0063] The dispersion curve is obtained from the above formula, with angular frequency on the horizontal axis and phase velocity on the vertical axis. Since different angular frequencies correspond to different numbers of modes, the angular frequency corresponding to the lowest number of modes needs to be selected as the optimal frequency for the excitation signal. The high-sensitivity mode corresponding to this angular frequency, i.e., the mode with a small change in angular frequency but a large change in phase velocity, can also be considered as the mode with a large slope in the dispersion curve; multiple modes with large slopes can be selected. The optimal frequency-mode combination obtained in this way does not have a fixed excitation signal frequency; typically, 20000 Hz - 2 × 10⁻⁶ Hz can be considered. 6 Dispersion curves in the Hz range.

[0064] Figure 2 The following describes the calculation process of the ultrasonic guided wave dispersion curve and sound field solution of the aero-engine blade according to the present invention: a three-dimensional curve coordinate system is established, and a three-dimensional model of the aero-engine blade is performed; the dispersion curve of the aero-engine blade is solved using semi-analytical finite element method; material properties are set, and the cross-sectional domain or line domain is discretized and divided into planar meshes, usually quadrilateral or triangular discretization; the characteristic solution of the control equation of the guided wave is solved, and information such as wavelength, phase velocity, group velocity, and mode is obtained from the characteristic solution, thus obtaining the dispersion curve.

[0065] Step 2: Set the obtained optimal frequency-mode combination as the excitation signal parameters; use the solid mechanics module as the physical field, set the excitation and receiving positions on the blade, and simulate in COMSOL Multiphysics software to obtain the ultrasonic guided wave detection signal in the aero-engine blade; perform multiple simulations to obtain the sound field distribution dataset.

[0066] The excitation and receiving positions can be arranged around the entire blade, 5 mm from the blade edge.

[0067] like Figure 2 As shown, a three-dimensional finite element modeling technique is used to simulate an aero-engine blade model containing typical defects such as holes, creep, scratches, tears, and fatigue cracks, and material properties are set. Boundary conditions for the wave equation are defined, and a perfect matching layer technique is used to suppress boundary reflections of the blade. The blade model is meshed, typically using quadrilateral or triangular discretization. Based on the optimal frequency-mode combination determined from the dispersion curve, the wave equation is solved in COMSOL Multiphysics after setting the excitation signal parameters, obtaining the ultrasonic guided wave sound field distribution in the aero-engine blade. Solving the wave equation is the starting point of the simulation in COMSOL Multiphysics, obtaining the ultrasonic guided wave sound field distribution in the aero-engine blade; the sound field distribution is the ultrasonic guided wave detection signal.

[0068] Multiple simulations were performed on aero-engine blades with typical defects to obtain simulation datasets. Experiments were conducted according to the simulation settings to establish experimental datasets. Simulation and experimental data were used to verify each other and evaluate the accuracy of the dispersion curve solution in the three-dimensional curvilinear coordinate system. Specifically, the correlation coefficient between the simulation dataset and the experimental dataset was calculated. If the correlation coefficient was greater than 0.75, the data obtained from the simulation and experiment were considered correct. The frequency-mode combination for defect detection was verified to be optimal. If it was not optimal, the angular frequency corresponding to the condition with fewer modes was selected as the optimal frequency of the excitation signal. The high-sensitivity mode corresponding to this angular frequency was then selected.

[0069] Step 3: Construct a dual-constraint aero-engine blade defect imaging network. Preprocess the ultrasonic guided wave detection signals in the dataset and train the defect imaging network. Input the preprocessed ultrasonic guided wave detection signals acquired in real time into the trained defect imaging network to establish a nonlinear mapping relationship from signal features to aero-engine blade defects, obtain defect information, and realize defect imaging of aero-engine blades.

[0070] like Figure 3 As shown, by preprocessing the signals in the simulation dataset and experimental dataset and constructing the imaging network, the accurate solution from the detection signal to the defect information can be achieved.

[0071] First, signal preprocessing is performed to obtain a standardized signal dataset. This dataset is then input into a dual-constraint aero-engine imaging network for imaging, which yields defect information. This process is called the nonlinear mapping from signal features to defects.

[0072] In the signal preprocessing stage, feature dimensionality reduction is performed on the hybrid simulation and experimental datasets: by extracting the first arrival wave of the ultrasonic guided wave signal, redundant signal segments are effectively eliminated, achieving data compression; subsequently, a fast Fourier transform is performed on the first arrival wave to obtain the amplitude spectrum and phase spectrum. The amplitude and phase corresponding to the excitation frequency are then identified in the amplitude and phase spectra, accurately extracting the characteristic amplitude and phase information corresponding to the excitation center frequency, constructing a three-dimensional feature matrix. The first and second dimensions represent the characteristic amplitude and phase, and the third dimension represents different modes. A zero-mean normalization algorithm is used to standardize the three-dimensional feature matrix, eliminating the dimensionality effects caused by differences in parameters of different detection devices, forming a standardized signal dataset. Feature dimensionality reduction reduces the amount of data input to the imaging network.

[0073] Based on the preprocessed standardized signal dataset, a dual-constraint aero-engine imaging network is designed. A multi-scale residual convolution module is constructed to mine the defect information in the characteristic frequencies and phases of multimodal ultrasonic guided wave signals. The multi-scale residual convolution module includes a multi-scale branch structure, which contains N parallel convolution branches. Each convolution branch uses a convolution kernel of a different scale to extract multi-scale features.

[0074] (5)

[0075] Where X is a three-dimensional feature matrix, X i For multi-scale features, Let k represent the convolution operation in the i-th branch. i The kernel size is usually an odd number, such as 3, 5, or 7, which are commonly used.

[0076] Multi-scale residual convolution is performed according to the following formula:

[0077] (6)

[0078] Here, Y represents the output defect information, * denotes the Hadamard product, Attention represents the attention mechanism, and Concat represents the concatenation operation along the channel dimension. Because the matrices are of the same shape, element-wise matrix multiplication using the Hadamard product is necessary.

[0079] A composite loss function strategy is adopted, calculating the error between the true label and the predicted defect as the loss, and using an L1 sparse regularization term to suppress noise interference in non-defect regions, thus minimizing the composite loss function L. total ,and:

[0080] (7)

[0081] Where L represents the error between the true label and the predicted defect, L sparse The regularization term L1 sparseness is represented by α, which represents the regularization strength and is typically chosen following the principle of "from small to large, range search". In defect detection tasks in industry and academia, the typical value range for α is [1e-5, 1e-2]. The more non-zero values ​​and the larger the numerical values ​​in the predicted defect information, the stronger the regularization term L1 sparseness becomes. sparse The larger the result, the greater the final composite loss function L. total The larger the value, the more it forces the model to make the predicted value of non-defective areas as close to 0 as possible.

[0082] By using the adaptive moment estimation algorithm to minimize the composite loss function, when the composite loss function is minimized, that is, when the error between the true label and the predicted defect and the L1 sparse regularization term are both small, it can be considered that the true label and the predicted defect are close, that is, the defect can be accurately predicted. This realizes the establishment of a nonlinear mapping relationship from signal features to aero-engine blade defects, and enables defect imaging of aero-engine blades.

[0083] Figure 3 The following describes the imaging process for aero-engine blade defects according to the present invention: The aero-engine blade defect imaging process involves: preprocessing the signals in the dataset and constructing an imaging network to achieve accurate solution from detection signals to defect information; in the signal preprocessing stage, feature dimensionality reduction is performed on the mixed simulation and experimental datasets; by extracting the first arrival wave of the multimodal ultrasonic guided wave signal, redundant signal segments are effectively eliminated, achieving data compression; subsequently, a fast Fourier transform is performed on the first arrival wave to accurately extract the characteristic frequencies and phase information corresponding to the excitation center frequency, constructing a three-dimensional feature matrix, where the first and second dimensions represent the characteristic frequencies and phases, and the third dimension represents different modes; a zero-mean normalization algorithm is used to standardize the three-dimensional feature matrix, eliminating the dimensional influence caused by differences in parameters of different detection equipment, forming a transferable standardized signal dataset; based on the preprocessed dataset, a dual-constraint aero-engine imaging network is designed; a multi-scale residual convolution module is constructed to mine defect information in the characteristic frequencies and phases of the multimodal ultrasonic guided wave signal; the multi-scale branch outputs are spliced ​​and the number of channels is adjusted through convolution, connected to the input residual, and multi-scale residual convolution is achieved through an attention mechanism. A composite loss function strategy is adopted, calculating the error between the true label and the predicted defect as the loss, and using an L1 sparse regularization term to suppress noise interference in non-defect regions, thus minimizing the composite loss function L. total Ultimately, a nonlinear mapping relationship between signal characteristics and aero-engine blade defects is directly established, enabling defect imaging of aero-engine blades.

[0084] Figure 4 The figure shows the aero-engine blade detection signal obtained from step two of this invention. The horizontal axis represents time, and the positive direction of the vertical axis represents the increasing sensor number. The time-domain sound field shown in the figure is the time-domain detection signal received by sensor 1 during the cyclic transmission of sensors 1-64. The reception delay increases with the distance between the transmitting and receiving sensors. Because aero-engine blades contain torsion and curved surfaces, and the sensor layout is asymmetrical, the delay in the received detection signal is also asymmetrical.

[0085] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for imaging defects in a dual-constraint aero-engine blades, characterized in that, The steps are as follows: Step 1: Perform 3D modeling of the aero-engine blade, calculate the ultrasonic guided wave dispersion curve of the aero-engine blade, and select the optimal frequency-mode combination corresponding to the excitation signal. Step 2: Set the obtained optimal frequency-mode combination as the parameters of the excitation signal; Using a solid mechanics module as the physical field, excitation and receiving positions were set on the blade to simulate and obtain the ultrasonic guided wave detection signal in the aero-engine blade; multiple simulations were performed to obtain the sound field distribution dataset. Step 3: Construct a dual-constraint aero-engine blade defect imaging network. Preprocess the ultrasonic guided wave detection signals in the dataset and train the defect imaging network. Input the real-time acquired ultrasonic guided wave detection signals into the trained defect imaging network to obtain defect information and realize defect imaging of aero-engine blades. The preprocessing involves performing feature dimensionality reduction on the hybrid dataset obtained from the simulation dataset and the experimental dataset to obtain a standardized signal dataset. Based on the standardized signal dataset, a dual-constraint aero-engine imaging network is designed, including a multi-scale residual convolution module. A composite loss function is used to train the dual-constraint aero-engine imaging network. The feature dimensionality reduction method is as follows: extract the first arrival wave of the ultrasonic guided wave detection signal and remove redundant signal segments; perform a fast Fourier transform on the first arrival wave to obtain the amplitude spectrum and phase spectrum; extract the feature amplitude and phase information corresponding to the excitation center frequency from the amplitude spectrum and phase spectrum, and construct a three-dimensional feature matrix, where the first and second dimensions represent the feature amplitude and phase, and the third dimension represents different modes; use a zero-mean normalization algorithm to standardize the three-dimensional feature matrix to obtain a standardized signal dataset. The multi-scale residual convolution module includes a multi-scale branch structure, which contains N parallel convolution branches. Each convolution branch uses convolution kernels of different scales to extract multi-scale features; and performs multi-scale residual convolution on the multi-scale features to obtain defect information. Simulation and experiment data are used to verify each other. The method of simulation-experiment mutual verification is as follows: calculate the correlation coefficient between the simulation dataset and the experimental dataset. If the correlation coefficient is greater than 0.75, the data obtained from simulation and experiment are considered to be correct. Verify whether the frequency-mode combination of defect detection is optimal. If it is not optimal, select the angular frequency corresponding to the condition with fewer modes as the optimal frequency of the excitation signal and select the mode corresponding to the optimal frequency. The multi-scale features Where X is a three-dimensional feature matrix, X i For multi-scale features, Let k represent the convolution operation in the i-th branch. i The kernel size; Defect information is obtained from the output using scale residual convolution. Where Attention represents the attention mechanism, and Concat represents the concatenation operation along the channel dimension. For Hadamah accumulation; The composite loss function is: ; Wherein, loss L represents the error between the true label and the predicted defect, loss L sparse α represents the regularization term for L1 sparseness, and α represents the regularization strength. By using the adaptive moment estimation algorithm to minimize the composite loss function, when the composite loss function is minimized, the error between the true label and the predicted defect information and the L1 sparse regularization term are both small. It is assumed that the true label and the predicted defect information are close, and the defect is accurately predicted, thus realizing defect imaging of aero-engine blades.

2. The dual-constraint aero-engine blade defect imaging method according to claim 1, characterized in that, The method for three-dimensional modeling of aero-engine blades is as follows: the aero-engine blades are scanned in three dimensions to obtain the point cloud coordinates of the blades; simulation is performed on three-dimensional finite element software, and the aero-engine blades are modeled in three dimensions using the point cloud coordinates to obtain three-dimensional curves.

3. The dual-constraint aero-engine blade defect imaging method according to claim 2, characterized in that, Let the coordinates of the three-dimensional curvilinear coordinate system be... Where x and y represent cross sections perpendicular to the blade axis, and s represents the blade axis; the position vector in the three-dimensional curve. Through Descartes orthogonal bases Represented as: ； Where r and d are the radius and helical step of the curve in the Cartesian coordinate system, and l is the length of the curve in the helical step. When the radius r = 0, the spiral centerline is the torsion centerline, and the three-dimensional curvilinear coordinate system is used... This indicates that for any position vector in the Cartesian coordinate system... Represented as: ； 4. The dual-constraint aero-engine blade defect imaging method according to any one of claims 1-3, characterized in that, The dispersion curve of aero-engine blades was solved using the semi-analytical finite element method.

5. The dual-constraint aero-engine blade defect imaging method according to claim 4, characterized in that, In the semi-analytical finite element method, the cross-sectional or linear domain is discretized and divided into a planar mesh. When the waveguide cross-section is symmetrical, the semi-analytical finite element formula for the one-dimensional mesh is the governing equation of the ultrasonic guided wave: ;in, For semi-analytical finite element mesh The nodal displacements; the matrix and Determined by the waveguide cross-section; Indicates wave number, Representation matrix The dimension is , and All are stiffness matrices. A symmetric matrix representing undamped motion; For each angular frequency The wave number k is obtained from the control equation of the ultrasonic guided wave, and thus the phase velocity is obtained: ； angular frequency With phase velocity c p The relationship is represented by a dispersion curve.

6. The dual-constraint aero-engine blade defect imaging method according to claim 5, characterized in that, Based on the convergence criterion for shear wave velocity in elastic materials, the maximum mesh length for semi-analytical finite element methods is: ;in, Maximum input frequency, It is a constant. Indicates the shear wave velocity; Since different angular frequencies correspond to different numbers of modes, the angular frequency corresponding to the lowest number of modes is selected as the optimal angular frequency for the excitation signal. The mode corresponding to the optimal angular frequency is the mode with the largest slope in the dispersion curve. The optimal frequency-mode combination is within 20000 Hz - 2 × 10⁻⁶. 6 Dispersion curves in the Hz range.

7. The dual-constraint aero-engine blade defect imaging method according to claim 5 or 6, characterized in that, A three-dimensional finite element model of an aero-engine blade with defects including holes, creep, scratches, tears, and fatigue cracks was simulated, and material properties were set. Boundary conditions for the wave equation were defined, and a perfect matching layer technique was used to suppress boundary reflections of the blade. The blade model was then meshed. The frequency and mode were directly obtained as parameters for the excitation signal. After setting the parameters of the excitation signal based on the optimal frequency-mode combination determined from the dispersion curve, the wave equation was solved to obtain the ultrasonic guided wave sound field distribution in the aero-engine blade. The sound field distribution is the ultrasonic guided wave detection signal. Multiple simulations were performed on defective aero-engine blades to obtain simulation datasets, and experiments were conducted according to the simulation settings to establish experimental datasets.

Images