Blade disc detection system based on flexible eddy current sensor

Abstract

The invention discloses a bladed disc detection system based on a flexible eddy current sensor, and the system employs rubber as a flexible substrate of a flexible array probe, can adapt to each complex region of the surface of a blisk, guarantees the fitting of the flexible array probe and the surface of the bladed disc to the greatest extent, and improves the detection precision. The flexible circuit board integrated with the array unit has higher deflection, and can ensure that the arrangement and the structure of the array coil are not damaged under the condition of simultaneously adapting to the curvature change of the leaf disc and the rubber; and the distances between the central point of the excitation coil or the detection coil on the outermost layer and the two sides of the flexible circuit board are respectively 6mm and 4mm, so that the detection blind area is smaller.

Description

Technical Field

[0001] The present application relates to the technical field of aero-engine integral blade disk detection, and in particular to a blade disk detection system based on a flexible eddy current sensor. Background Art

[0002] The aeroengine blisk is a novel structural component designed to meet the demands of high-performance aeroengines. It utilizes a single-piece manufacturing process that integrates the rotor blades and disk, eliminating the traditional connection process involving mortise and tenon joints and locking mechanisms. This reduces structural weight and part count, minimizes airflow loss, improves starting efficiency, and simplifies engine structure. Blisks are typically manufactured from titanium or nickel-based alloys and are primarily used in fans and high-pressure compressors, conveying airflow and bearing mechanical loads. The rotating blades exert work on the airflow, guiding it in its designed direction. The centrifugal force generated by the blades' rotation is transmitted to the disk, which, as the primary load-bearing structure, transfers the load to the engine's main shaft. With blades and disk cast as one piece, blisks are subject to more complex and variable loads during aeroengine operation. Furthermore, stress concentrations occur at the geometrically disjointed locations at the blade roots. In summary, blades subjected to alternating stress and aerodynamic bending stress at high rotational speeds are prone to fatigue cracking. The mismatch in thermal expansion coefficients between the blade surface thermal barrier coating and the base material creates interfacial stress that can easily cause the coating to peel. Furthermore, inhaled external particles or internal debris (such as runway debris, bird strikes, hail, and coating fragments from the front-stage compressor blades) can impact the blades, causing notches or pits and inducing cracks. The safety of an aero-engine blisk is a crucial factor in ensuring its safety, reliability, and efficiency. Its design, manufacturing, and defect detection directly determine the safety, reliability, and comfort of the aircraft.

[0003] Eddy current testing (PE) technology is a non-destructive testing method based on the principle of electromagnetic induction. It uses the law of electromagnetic induction to pass an alternating current of a certain frequency through the detection coil, generating an alternating magnetic field (primary magnetic field) around the coil. When the conductor (structural component) to be tested is placed in the magnetic field, a circular current (eddy current) will be generated on the surface of the conductor. The eddy current itself generates a reverse magnetic field (secondary magnetic field), which is superimposed on the original magnetic field, thereby changing the impedance of the coil (voltage / current phase difference). Defects on the surface or near the surface of the conductor (such as cracks, corrosion, inclusions, etc.) will change the eddy current path, causing abnormal secondary magnetic fields and changes in the coil impedance, thereby inferring the location and size of the defect. Due to the characteristics of non-contact detection, high sensitivity, low cost, and easy automation of eddy current testing, it is widely used in fields such as rail transportation and aerospace.

[0004] CN202510036130.9 discloses a flexible array eddy current detection probe and a method of use, which aims to solve the problem that traditional array eddy current detection devices cannot handle complex test piece surfaces well. The detection device includes: a flexible silicone base, a flexible printed circuit board, a TR probe unit, an eddy current detector and a processing terminal. The flexible silicone pad is used to provide flexible support, ensure that the detection surface fits the surface to be tested, and reduce the lift-off effect; the flexible printed circuit board integrates the TR probe unit while maintaining the flexibility of the overall structure; the TR probe unit consists of an excitation coil and a detection coil, and the two work in a differential manner; the eddy current detector is used to generate an excitation current and collect differential signals, and the processing terminal processes the differential signals through an algorithm to identify the crack location and size, etc.

[0005] The detection method is as follows: first, a flexible substrate is attached to the surface of a turbine blade to ensure that the TR probe unit is in contact with the surface to be inspected; a sinusoidal current is input into the excitation coil through an eddy current detector to generate a time-varying magnetic field, thereby generating an induced eddy current field in the blade; when the eddy current field is in a defect-free area, it is evenly distributed, and the detection coil outputs a symmetrical signal; when the eddy current field is in a defective area, the crack cuts the magnetic flux lines, causing the eddy current field to be distorted, and the detection coil outputs an abnormal differential signal; the eddy current instrument synchronously collects the differential voltage signals from the four TR probes, transmits them to a processing terminal through leads, and finally determines whether there is a defect, as well as the defect boundary and size, by calculating the sensitivity.

[0006] However, eddy current nondestructive testing technology is widely used in the aviation field to detect defects in various metal structures. However, the integral blade disk of an aircraft engine has complex and changeable free-form surface characteristics. Traditional rigid probes not only have problems such as poor fitting and excessive blind spots, but may also cause further damage to the integral blade disk during the inspection process. Traditional flexible probes with silicone as a base are easily difficult to use or cannot be used due to silicone denaturation. Traditional flexible probes with silicone as a base may form a sticky layer or cause silicone to peel off under the influence of certain factors (such as ambient temperature and humidity, silicone manufacturing process, and contact with chemical reagents used to protect the surface of the workpiece being inspected), resulting in poor inspection results and even contamination of the surface of the workpiece being inspected. Summary of the Invention

[0007] The purpose of the present application is to provide a blade disk detection system based on a flexible eddy current sensor to solve the problem in the above-mentioned background technology that the flexible probe of the eddy current detection device with silicone as the base is easily difficult to use due to the denaturation of silicone.

[0008] To achieve the above objectives, this application provides the following technical solutions:

[0009] A flexible eddy current detection system based on a blade disk curved surface is applied to a flexible eddy current detection device based on a blade disk curved surface. The detection system includes:

[0010] an FPGA module configured to determine a working frequency of the eddy current detector;

[0011] a signal processing module configured to extract a blade defect feature from a detection signal obtained by the eddy current detector after noise reduction, and to perform three-dimensional reconstruction on the blade defect feature; the blade defect feature includes defect length, width, depth, and angle;

[0012] a detection report generation module configured to generate a blade defect detection report, the blade defect detection report including a defect distribution heat map and a quantitative damage index.

[0013] In one embodiment, the flexible eddy current detection device based on a blade surface includes a mechanical arm, a flexible array probe, and an eddy current detector. The mechanical arm carries the flexible array probe, which is used to obtain a defect signal of the blade. The eddy current detector is arranged on one side of the mechanical arm and is used to obtain detection data of the flexible array probe and generate a detection report from the detection data.

[0014] The flexible array probe includes a rubber base, a flexible circuit board, and an array unit. The flexible circuit board is mounted on the rubber base, and the array unit is integrated on the flexible circuit board. The array unit includes an excitation coil and a detection coil. The center points of the outermost excitation coil or detection coil are 6 mm and 4 mm away from the two edges of the flexible circuit board, respectively.

[0015] In one embodiment, the coils in the array unit are connected in a bridge mode, so that the two coils in each channel of the array unit are excitation coils and detection coils.

[0016] In one embodiment, the center points of the excitation coil and the detection coil in each channel are 5 mm apart, so that the magnetic field can cover the entire detection area without crosstalk.

[0017] In one embodiment, the rubber elastic modulus of the rubber base is 0.6 Mpa-6.1 Mpa, and the use temperature is -40℃-125℃.

[0018] In one embodiment, the FPGA module includes a working frequency sending unit, a signal receiving unit, a working frequency determining unit, a working frequency adjusting unit, and a blade material feature unit.

[0019] The working frequency sending unit is configured to send an initial working frequency or an adjusted working frequency to the flexible array probe.

[0020] The signal receiving unit is configured to receive a feedback signal of the flexible array probe. The feedback signal includes signal-to-noise ratio, sensitivity, and phase consistency.

[0021] The operating frequency judgment unit is used to judge whether the initial operating frequency f0 or the adjusted operating frequency f is the optimal operating frequency of the eddy current detector according to the feedback signal;

[0022] The operating frequency adjustment unit determines the adjusted operating frequency f according to the detected signal-to-noise ratio, sensitivity, phase consistency and corresponding weight coefficients, where f=f0+Δf, Δf: frequency adjustment step size;

[0023] The blade disk material characteristic unit is used to determine the weight coefficients corresponding to the signal-to-noise ratio, sensitivity, and phase consistency according to the detected blade disk material.

[0024] In one embodiment, the frequency adjustment step Δf is calculated as follows: Δf=K×Δfmax, where Δfmax is the maximum frequency adjustment amount.

[0025] In one embodiment, the signal processing module includes a signal preprocessing unit, a deep learning denoising unit, and a three-dimensional reconstruction unit;

[0026] The signal preprocessing unit is used to normalize the detection signal, eliminate the dimensional differences between different features, and use a Hamming window to perform frame processing on the normalized detection signal, automatically match the noise type label, and send it to the deep learning denoising unit;

[0027] The deep learning denoising unit is used to output the denoised defect features based on the detection signal with the noise type label and the preset material parameters;

[0028] The three-dimensional reconstruction unit is used to convert abstract defect features into an intuitive 3D model, clarify the depth, length, and volume parameters of the defect, and accurately restore the geometric shape, position, size, and distribution of the defect in three-dimensional space.

[0029] In one embodiment, the deep learning denoising unit includes a maximum pooling layer subunit, a first SE attention subunit, a material feature subunit, a deconvolution upsampling subunit, a skip connection high-frequency detail supplementation subunit, and a defect feature acquisition subunit;

[0030] The maximum pooling layer subunit is used to extract the most significant features in the region while reducing the data dimension. The most significant features include noise features and defect features.

[0031] The first SE attention sub-unit is used to enhance the high-level semantic feature response of the defect feature sensitive frequency band through channel weight allocation; the high-level semantic features include defect features assigned high weights and noise features assigned low weights;

[0032] The material feature subunit is used to map the pre-stored material parameters into feature weight vectors through the fully connected layer, and multiply them channel by channel with the high-level semantic features output by the first SE attention subunit to achieve the fusion of material physical properties and signal features, remove material-irrelevant noise features, and highlight defect-sensitive features;

[0033] The deconvolution upsampling subunit is used to reconstruct the macroscopic outline of the defect using the weighted defect-sensitive features;

[0034] The skip connection supplements the high-frequency detail subunit, and is used to perform channel splicing of the shallow features of the corresponding level of the encoder and the decoder features using the skip connection, thereby retaining the high-frequency details of the defect-sensitive features on the macroscopic outline of the reconstructed defect to obtain the reconstructed defect features;

[0035] The defect feature acquisition subunit is used to map the reconstructed defect feature into a denoised defect feature signal.

[0036] In one embodiment, the deep learning denoising unit further includes a second SE attention sub-unit, which is used for feature channel weighting, enhancing reconstruction defect features, and suppressing noise features.

[0037] Compared with the prior art, the present invention has the following advantages:

[0038] The above-mentioned flexible eddy current detection system based on the curved surface of the blade disk adopts rubber as the flexible substrate of the flexible array probe, which can adapt to various complex areas of the surface of the entire blade disk, and ensure the maximum fit between the flexible array probe and the surface of the blade disk. The flexible circuit board with integrated array unit has high flexibility, which can ensure that the arrangement and structure of the array coil are not damaged while adapting to the changes in the curvature of the blade disk and rubber at the same time; and the center point of the outermost excitation coil or detection coil is 6mm and 4mm away from the two sides of the flexible circuit board respectively, so that the blind area of ​​detection is smaller. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 This is a structural schematic diagram of a flexible eddy current detection device based on the curved surface of a blade disk in this application;

[0040] Figure 2 for Figure 1 A schematic diagram of the flexible array probe structure of a flexible eddy current detection device based on the curved surface of a blade disk in this application;

[0041] Figure 3 This is a structural schematic diagram of a flexible eddy current detection system based on the curved surface of a blade disk in this application.

[0042] Fig. 1 is a schematic diagram of a flexible eddy current testing device based on a blade surface, comprising a mechanical arm 1, a flexible array probe 2, a rubber base 21, a flexible circuit board 22, an array unit 23, an excitation coil 231, a detection coil 232, an eddy current tester 3, an FPGA module 100, a signal processing module 200, and a detection report generation module 300. DETAILED DESCRIPTION

[0043] In order to make the above objectives, features and advantages of the present application more apparent, specific embodiments of the present application will be described in detail below with reference to the accompanying drawings. In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present application. However, the present application can be practiced in many different ways from those described herein without departing from the scope of the present application, and it is understood that similar improvements can be made by those skilled in the art without departing from the spirit of the present application, and therefore the present application is not limited to the specific embodiments disclosed below.

[0044] It should be noted that when an element is referred to as being "connected" to another element, it can be directly connected to the other element or can exist with an intervening element. In contrast, when an element is referred to as being "directly connected" to another element, there is no intervening element.

[0045] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs. The terminology used in the description of the present application herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the present application. As used herein, the term "and / or" includes any and all combinations of one or more of the associated listed items.

[0046] As shown in Figure 1 , Figure 2 A flexible eddy current testing device based on a blade surface, comprising a mechanical arm 1, a flexible array probe 2, and an eddy current tester 3, the mechanical arm 1 carrying the flexible array probe 2, the flexible array probe 2 being used to acquire defect signals of the blade, the eddy current tester 3 being disposed on one side of the mechanical arm 1, used to acquire detection data of the flexible array probe 2, and generate a detection report from the detection data.

[0047] The flexible array probe 2 comprises a rubber base 21, a flexible circuit board 22, and an array unit 23, the flexible circuit board 22 being mounted on the rubber base 21, and the array unit 23 being integrated on the flexible circuit board 22; the array unit 23 comprises an excitation coil 231 and a detection coil 232; the center point of the outermost excitation coil 231 or detection coil 232 is 6 mm and 4 mm away from the two edges of the flexible circuit board 22, respectively.

[0048] The flexible circuit board 22 is a highly reliable and flexible printed circuit board made of polyester film or polyimide and formed by etching circuits on copper foil. The array unit 23 is integrated on the flexible circuit board 22 and has high flexibility. While integrating the array unit 23 with the leads, it can be moved and extended arbitrarily in three-dimensional space and can adapt to the curvature changes of the entire blade disk.

[0049] The coils in the array unit 22 are connected in a bridge manner, so that the two coils in each channel of the array unit 22 serve as an excitation coil 231 and a detection coil 232 for each other;

[0050] The center distance between the excitation coil 231 and the detection coil 232 of each channel is 5 mm, so that the magnetic field can cover the entire detection area without crosstalk;

[0051] The array units 23 are 6mm and 4mm away from the edges of the flexible printed circuit board 22, respectively, meeting the requirements for nondestructive testing of the entire blade disk. The flexible array probes 2 are arranged in a regular pattern to ensure coordinated operation during testing. The center points of the excitation coil 231 and detection coil 232 in each channel are 5mm apart. This ensures that the magnetic field covers the entire testing area without crosstalk, ensuring no missed detections.

[0052] The elastic modulus of the rubber in the rubber base 21 is 0.6 MPa-6.1 MPa, and the operating temperature is -40°C-125°C.

[0053] The rubber elastic modulus of the rubber substrate 21 is 0.6Mpa-6.1Mpa, and its operating temperature is between -40°C and 125°C, which meets the requirements for detecting the out-of-position dislocation of an aircraft engine's integral blade disc. By creating a rubber substrate 21 that is both flexible and compressible as a flexible substrate for the flexible circuit board 22, a certain degree of support can be provided for the flexible circuit board 22, ensuring better adhesion between the flexible circuit board 22 and the complex curved surface of the integral blade disc, reducing the lift-off effect, and enhancing the magnetic field coupling efficiency. Compared to silicone, rubber is more oil-resistant and less likely to react with the chemical substances used to protect the surface of the integral blade disc. However, silicone tends to produce sticky substances on its surface during operation, which can reduce sensor reliability or cause sensor wear.

[0054] When inspecting the surface of the blade disk, the robotic arm 1 controls the flexible array probe 2 to fit the surface of the blade disk and sends the detection signal to the eddy current detector 3. The eddy current detector 3 obtains the defect characteristics of the blade disk based on the detection signal and generates a blade disk defect detection report.

[0055] In this way, a flexible eddy current detection system based on the curved surface of the blade disk is provided. The present application adopts rubber as the flexible substrate of the flexible array probe 2, which can adapt to various complex areas of the surface of the entire blade disk, and ensure that the flexible array probe 2 fits the surface of the blade disk to the greatest extent. The flexible circuit board 22 with integrated array unit 23 has high flexibility, which can ensure that the arrangement and structure of the array coil are not damaged while adapting to the changes in the curvature of the blade disk and rubber at the same time; and the center point of the outermost excitation coil 231 or the detection coil 232 is 6 mm and 4 mm away from the two sides of the flexible circuit board 22 respectively, so that the blind area of ​​detection is smaller.

[0056] like Figure 3 As shown, a flexible eddy current detection system based on a blade disk curved surface is applied to the detection system in the flexible eddy current detection device based on a blade disk curved surface, and the detection system includes:

[0057] FPGA module 100, used to determine the operating frequency of the eddy current detector 3;

[0058] The signal processing module 200 is used to extract the defect features of the blade disk after de-noising the detection signal obtained by the eddy current detector 3, and perform three-dimensional reconstruction of the defect features of the blade disk; the defect features of the blade disk include defect length, width, depth, and angle;

[0059] The inspection report generation module 300 is used to generate a blade disk defect inspection report, wherein the blade disk defect inspection report includes a defect distribution heat map and a quantitative damage index.

[0060] In one embodiment, the FPGA module 100 includes an operating frequency sending unit, a signal receiving unit, an operating frequency determining unit, an operating frequency adjusting unit, and a blade disk material characteristic unit;

[0061] The operating frequency sending unit is used to send the initial operating frequency f0 or adjust the operating frequency f to the flexible array probe 2;

[0062] The signal receiving unit is used to receive the feedback signal of the flexible array probe 2; the feedback signal includes signal-to-noise ratio, sensitivity, and phase consistency;

[0063] The operating frequency judgment unit is used to judge whether the initial operating frequency f0 or the adjusted operating frequency f is the optimal operating frequency of the eddy current detector according to the feedback signal;

[0064] The operating frequency adjustment unit determines the adjusted operating frequency f according to the detected signal-to-noise ratio, sensitivity, phase consistency and corresponding weight coefficients, where f=f0+Δf, Δf: frequency adjustment step size;

[0065] The blade disk material characteristic unit is used to determine the weight coefficients corresponding to the signal-to-noise ratio, sensitivity, and phase consistency according to the detected blade disk material.

[0066] The calculation formula of the frequency adjustment step Δf is: Δf=K×Δfmax, where Δfmax is the maximum frequency adjustment amount.

[0067] The operating frequency adjustment of traditional eddy current detectors 3 relies on manual experience, which is inefficient and easy to miss the optimal frequency parameters. This application dynamically optimizes the operating frequency by receiving and analyzing the real-time feedback signal (signal-to-noise ratio, sensitivity and phase consistency).

[0068] A preset initial operating frequency f0 is set in the eddy current detector 3; the operating frequency sending unit of the eddy current detector 3 applies an excitation current with a preset initial operating frequency f0 to the excitation coil 231; the excitation current generates an eddy current field on the surface of the blade disk, and the detection coil 232 receives a feedback signal caused by the eddy current field and sends the feedback signal to the signal receiving unit; the signal receiving unit sends the feedback signal to the operating frequency determination unit;

[0069] The operating frequency judgment unit judges whether the feedback signal exceeds a preset threshold. If so, the initial operating frequency f0 is determined to be the optimal operating frequency; the initial operating frequency f0 is determined to be the operating frequency of the eddy current detector 3 .

[0070] If the feedback signal is less than the preset threshold, the feedback signal is sent to the operating frequency adjustment unit;

[0071] The operating frequency adjustment unit obtains the signal-to-noise ratio, sensitivity, and phase consistency of the detected relevant parameters; the signal-to-noise ratio (SNR) = 10log10(Psignal / Pnoise), SNR ≥ 4 is a high SNR, which is determined to be a clear defect signal feature, SNR = 2-4 is a medium SNR, which is determined to be a high probability of a defect signal, and SNR = 1-2 is a low SNR, which may indicate an uneven surface of the blade disk material or a pseudo-defect.

[0072] Sensitivity (η) = △Vdefect / Vbaseline * 100%, where Vbaseline is the baseline signal amplitude in the defect-free area, and Vdefect is the change in signal amplitude in the defect area. The grading standard can be referenced as follows: η ≥ 10% is high sensitivity, suitable for microcrack detection; η = 5% to 10% is medium sensitivity, suitable for pit detection; η < 5% is low sensitivity, which requires combination with other indicators for judgment.

[0073] Phase consistency = 1-(π / σφ), where σφ is the standard deviation (radians) of the phase angle of the defect area, and the phase consistency ranges from 0 (complete inconsistency) to 1 (complete consistency). Phase consistency ranges from 0 (complete inconsistency) to 1 (complete consistency). The threshold setting can be referenced as follows: φcoh ≥ 0.8 is high consistency, that is, the phase is concentrated and the defect characteristics are obvious; φcoh = 0.5 ~ 0.8 is medium consistency, which needs to be judged in combination with the signal amplitude; φcoh < 0.5 is low consistency, which may be noise or a pseudo-defect. The detected signal-to-noise ratio, sensitivity, and phase consistency are normalized; the method for normalizing the signal-to-noise ratio, sensitivity, and phase consistency of the detected feedback signal is as follows:

[0074]

[0075] SNRnorm: signal-to-noise ratio normalization value, range: [0,1]; SNR: signal-to-noise ratio;

[0076] For example, if SNR=1, then SNRnorm=0.25;

[0077]

[0078] For example, if η = 15%, then ηnorm = 1;

[0079] ηnorm: sensitivity normalization value, range: [0,1]; η: sensitivity;

[0080] φcoh=1-(π / σφ), φcoh: phase consistency; σφ: phase standard deviation. Phase consistency does not require normalization.

[0081] For example, φcoh can be directly output as 0.6.

[0082] Because different blade disk materials will affect the signal-to-noise ratio, sensitivity, and phase consistency during detection. For example, high conductivity materials (copper, aluminum) may rely more on SNR (signal-to-noise ratio) because noise can easily mask small defects. Should be higher; magnetic materials (steel) may be more dependent on phase consistency (due to hysteresis effect affecting the phase), in this case Therefore, when adjusting the operating frequency, the influence of the blade disk material on the operating frequency needs to be considered. Therefore, the weight coefficients of the signal-to-noise ratio, sensitivity, and phase consistency are adjusted accordingly through the blade disk material characteristic unit to ensure the accuracy of the operating frequency of the eddy current detector 3.

[0083] According to the blade disk material to be tested, the blade disk material characteristic unit determines the weight coefficients of the signal-to-noise ratio, sensitivity, and phase consistency and sends them to the working frequency adjustment unit, which calculates the comprehensive adjustment coefficient K;

[0084] Signal-to-noise ratio weight coefficient: Sensitivity weight coefficient: Phase consistency weight coefficient:

[0085] The output frequency adjustment step △f is determined according to the detected signal-to-noise ratio, sensitivity, phase consistency and the corresponding weight coefficient; the calculation formula of the frequency adjustment step △f is obtained according to the preset rules of the working frequency adjustment unit: △f=K×△fmax, △fmax: maximum frequency adjustment amount.

[0086] For example, aircraft engine blade disk (titanium alloy) inspection:

[0087] f0 = 100 kHz, measured SNRnorm = 0.3, ηnorm = 1, φcoh = 0.75; titanium alloy weight: SNR (0.4), η (0.3), φ (0.3);

[0088] K=0.4×0.3+0.3×1.0+0.3×0.75=0.705

[0089] Δfmax=100kHz,K=0.705,then Δf=70.5kHz;

[0090] The adjusted operating frequency f = f0 + Δf is determined; if f0 = 100 kHz, then f = 100 + 70.5 = 170.5 kHz. The operating frequency transmitting unit of the eddy current detector 3 applies an excitation current to the excitation coil with an adjusted operating frequency of f. The signal receiving unit receives the feedback signal and sends it to the operating frequency determination unit. The operating frequency determination unit determines whether the feedback signal exceeds a preset threshold. If so, the adjusted operating frequency f is determined to be the optimal operating frequency. The adjusted operating frequency f is determined to be the operating frequency of the eddy current detector 3. By scientifically adjusting the operating frequency of the eddy current detector 3, the reliability and efficiency of eddy current testing can be significantly improved, avoiding missed or false detections.

[0091] In one embodiment, the signal processing module 200 includes a signal preprocessing unit, a deep learning denoising unit, and a three-dimensional reconstruction unit;

[0092] The signal preprocessing unit is used to normalize the detection signal, eliminate the dimensional differences between different features, and use a Hamming window to frame the normalized detection signal, automatically match the noise type label, and send it to the deep learning denoising unit; the window length of the Hamming window is 256 to 512 sampling points; the overlap rate of adjacent windows is 30% to 50%; the continuous signal is divided into short-time overlapping frames (for example, 256 to 512 sampling points per frame, and the overlap rate is 30% to 50%). After framing, the amount of data per frame is small and can be calculated in parallel (such as GPU acceleration), significantly reducing the time consumption of single processing.

[0093] The deep learning denoising unit is used to output the noise-reduced defect characteristics based on the detection signal with the noise type label and the preset material parameters; the deep learning denoising unit is a deep learning denoising model trained by combining synthetic data and real data;

[0094] The deep learning denoising unit includes the maximum pooling layer subunit, the first SE attention subunit, the material feature subunit, the deconvolution upsampling subunit, the skip connection high-frequency detail supplement subunit, and the defect feature acquisition subunit;

[0095] The maximum pooling layer subunit is used to extract the most significant features in the region while reducing the data dimension. The most significant features include noise features and defect features.

[0096] The maximum pooling layer subunit is used to extract the most significant features in the detection signal with noise type labels while reducing the data dimension. The most significant features include noise features and defect features; reducing data dimension: reducing the amount of calculation and memory usage; enhancing robustness: retaining the most significant local response and suppressing small fluctuations. Significance of noise features: high-frequency noise (such as impulse noise) or strong environmental interference may appear locally as "significant" peaks with sudden amplitude changes, which are retained by the maximum pooling layer subunit. Significance of defect features: signal mutations (such as amplitude drops and phase jumps) caused by real defects (such as cracks and pits) will also be captured by the maximum pooling layer subunit. For example, the amplitude of the eddy current signal in the crack area is significantly lower than the background, and the maximum pooling layer subunit will retain these abnormally low value points.

[0097] The first SE attention subunit is used to enhance the high-level semantic feature response of the defect feature sensitive frequency band through channel weight allocation; the high-level semantic features include defect features assigned high weights and noise features assigned low weights;

[0098] High weights are assigned to frequency bands sensitive to defect characteristics (such as specific resonance frequencies caused by cracks), while low weights are assigned to frequency bands dominated by noise characteristics (such as environmental electromagnetic interference). When outputting signals, frequency bands sensitive to defect characteristics will be retained.

[0099] The material feature subunit is used to map the pre-stored material parameters into feature weight vectors through the fully connected layer, and multiply them channel by channel with the high-level semantic features output by the first SE attention subunit to achieve the fusion of material physical properties and signal features, remove material-irrelevant noise features, and highlight defect-sensitive features; high-level semantic features and material parameters (conductivity σ, magnetic permeability μ) are fused through the fully connected layer to generate physically enhanced semantic features. For example, if the current material is aluminum alloy (σ=3.5×10 7 S / m), the features automatically enhance the ability to suppress high-frequency noise. High-level features can distinguish between signal changes caused by defects and irrelevant noise. For example, high-frequency noise appears sparsely activated in high-level features, while crack features exhibit a continuous activation pattern. This amplifies high-frequency defect features and weakens low-frequency noise features, making it easier for subsequent networks to identify small cracks.

[0100] The deconvolution upsampling subunit is used to reconstruct the macroscopic outline of the defect using the weighted defect-sensitive features;

[0101] High-level features contain global information about the defect (such as the main crack direction and the overall shape of the pore), but lack spatial details. The deconvolution upsampling subunit maps these abstract semantics into a rough defect geometry outline through deconvolution upsampling by expanding the spatial dimension. Because the input is weighted features, the reconstructed outline naturally conforms to material properties (such as the sharp edge of a metal crack or the layered propagation pattern of a composite material).

[0102] The skip connection supplements the high-frequency detail subunit, and is used to perform channel splicing of the shallow features of the corresponding level of the encoder and the decoder features using the skip connection, thereby retaining the high-frequency details of the defect-sensitive features on the macroscopic outline of the reconstructed defect to obtain the reconstructed defect features;

[0103] The high-frequency detail defect features are complemented on the macroscopic contour reconstructed according to the corrected defect features output by the deconvolution upsampling subunit, generating reconstructed defect features that conform to the physical laws of the material (macroscopic) and retain the real signal details (microscopic).

[0104] In one embodiment, the deep learning denoising unit further includes a second SE attention sub-unit, which is used for feature channel weighting, enhancing reconstruction defect features, and suppressing noise features.

[0105] Through the SE mechanism, the network automatically learns which channels contribute most to defect reconstruction (such as frequency bands or spatial patterns related to cracks and pores) and assigns higher weights to make the reconstructed defects clearer.

[0106] The defect feature acquisition subunit is used to map the reconstructed defect feature into a denoised defect feature signal.

[0107] In this way, through the defect feature acquisition sub-unit, the model completes the conversion from "abstract features" to "physical signals", so that the denoised defect feature signals can be directly used for engineering decision-making (such as defect classification and life prediction).

[0108] The three-dimensional reconstruction unit is used to convert abstract defect features into an intuitive 3D model, clarify the depth, length, and volume parameters of the defect, and accurately restore the geometric shape, position, size, and distribution of the defect in three-dimensional space.

[0109] The above is only a preferred specific implementation method of the present application, but the scope of protection of the present application is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes based on the technical solution and inventive concept of the present application within the technical scope disclosed in the present application, and they should be covered by the scope of protection of the present application.

Claims

1. A flexible eddy current detection system based on the curved surface of a blade disk, characterized in that: Applied to a flexible eddy current detection device based on the curved surface of a blade disk, the detection system includes: An FPGA module, used for determining the operating frequency of the eddy current detector; A signal processing module is used to extract the defect features of the blade disk after noise reduction of the detection signal obtained by the eddy current detector, and to perform three-dimensional reconstruction of the defect features of the blade disk; the defect features of the blade disk include defect length, width, depth, and angle; The detection report generation module is used to generate a blade disk defect detection report, wherein the blade disk defect detection report includes a defect distribution heat map and a quantitative damage index.

2. The flexible eddy current detection system based on the curved surface of a blade disk according to claim 1, characterized in that: The flexible eddy current detection device based on the blade disk curved surface includes a mechanical arm, a flexible array probe and an eddy current detector. The mechanical arm is equipped with the flexible array probe, and the flexible array probe is used to obtain defect signals of the blade disk. The eddy current detector is arranged on one side of the mechanical arm, and is used to obtain detection data of the flexible array probe and generate a detection report based on the detection data. The flexible array probe includes a rubber base, a flexible circuit board, and an array unit. The flexible circuit board is mounted on the rubber base, and the array unit is integrated on the flexible circuit board. The array unit includes an excitation coil and a detection coil. The center point of the outermost excitation coil or the detection coil is 6 mm and 4 mm away from the two sides of the flexible circuit board, respectively.

3. The flexible eddy current detection system based on the blade disk curved surface according to claim 2, characterized in that: The coils in the array unit are connected in a bridge-type manner, so that the two coils in each channel of the array unit serve as excitation coils and detection coils for each other.

4. The flexible eddy current detection system based on the blade disk curved surface according to claim 3, characterized in that: The distance between the center point of the excitation coil and the detection coil of each channel is 5 mm, so that the magnetic field can cover the entire detection area without crosstalk.

5. The flexible eddy current detection system based on the blade disk curved surface according to claim 2, characterized in that: The elastic modulus of the rubber in the rubber base is 0.6 MPa-6.1 MPa, and the operating temperature is -40°C-125°C.

6. The flexible eddy current detection system based on the blade disk curved surface according to claim 1, characterized in that: The FPGA module includes an operating frequency sending unit, a signal receiving unit, an operating frequency judging unit, an operating frequency adjusting unit, and a blade disk material characteristic unit; The operating frequency sending unit is used to send an initial operating frequency or adjust the operating frequency to the flexible array probe; The signal receiving unit is used to receive the feedback signal of the flexible array probe; the feedback signal includes signal-to-noise ratio, sensitivity, and phase consistency; The operating frequency judgment unit is used to judge whether the initial operating frequency f0 or the adjusted operating frequency f is the optimal operating frequency of the eddy current detector according to the feedback signal; The operating frequency adjustment unit determines the adjusted operating frequency f according to the detected signal-to-noise ratio, sensitivity, phase consistency and corresponding weight coefficients, where f=f0+Δf, Δf: frequency adjustment step size; The blade disk material characteristic unit is used to determine the weight coefficients corresponding to the signal-to-noise ratio, sensitivity, and phase consistency according to the detected blade disk material.

7. The flexible eddy current detection system based on the blade disk curved surface according to claim 6, characterized in that: The calculation formula of the frequency adjustment step Δf is: Δf=K×Δfmax, where Δfmax is the maximum frequency adjustment amount.

8. The flexible eddy current detection system based on the blade disk curved surface according to claim 1, characterized in that: The signal processing module includes a signal preprocessing unit, a deep learning denoising unit, and a three-dimensional reconstruction unit; The signal preprocessing unit is used to normalize the detection signal, eliminate the dimensional differences between different features, and use a Hamming window to perform frame processing on the normalized detection signal, automatically match the noise type label, and send it to the deep learning denoising unit; The deep learning denoising unit is used to output the denoised defect features based on the detection signal with the noise type label and the preset material parameters; The three-dimensional reconstruction unit is used to convert abstract defect features into an intuitive 3D model, clarify the depth, length, and volume parameters of the defect, and accurately restore the geometric shape, position, size, and distribution of the defect in three-dimensional space.

9. The flexible eddy current detection system based on the blade disk curved surface according to claim 8, characterized in that: The deep learning denoising unit includes a maximum pooling layer subunit, a first SE attention subunit, a material feature subunit, a deconvolution upsampling subunit, a skip connection high-frequency detail supplementation subunit, and a defect feature acquisition subunit; The maximum pooling layer subunit is used to extract the most significant features in the region while reducing the data dimension. The most significant features include noise features and defect features. The first SE attention sub-unit is used to enhance the high-level semantic feature response of the defect feature sensitive frequency band through channel weight allocation; the high-level semantic features include defect features assigned high weights and noise features assigned low weights; The material feature subunit is used to map the pre-stored material parameters into feature weight vectors through the fully connected layer, and multiply them channel by channel with the high-level semantic features output by the first SE attention subunit to achieve the fusion of material physical properties and signal features, remove material-irrelevant noise features, and highlight defect-sensitive features; The deconvolution upsampling subunit is used to reconstruct the macroscopic outline of the defect using the weighted defect-sensitive features; The skip connection supplements the high-frequency detail subunit, and is used to perform channel splicing of the shallow features of the corresponding level of the encoder and the decoder features using the skip connection, thereby retaining the high-frequency details of the defect-sensitive features on the macroscopic outline of the reconstructed defect to obtain the reconstructed defect features; The defect feature acquisition subunit is used to map the reconstructed defect feature into a denoised defect feature signal.

10. The flexible eddy current detection system based on the blade disk curved surface according to claim 9, characterized in that: The deep learning denoising unit also includes a second SE attention subunit, which is used for feature channel weighting, enhancing reconstruction defect features, and suppressing noise features.

Images