Double-probe film thickness and defect collaborative decoupling method

Abstract

The invention discloses a double-probe film thickness and defect collaborative decoupling method. The method comprises the steps that microwave reflection signals of a thermal barrier coating are synchronously collected through a coaxial resonance probe and a CSRR resonance ring probe; the microwave reflection signal comprises reflection coefficient amplitude, reflection coefficient phase and resonant frequency; constructing a frequency deviation model based on the microwave reflection signal; training the dual-channel deep learning network based on the frequency deviation model to obtain a trained dual-channel deep learning network; and obtaining film thickness information and a defect positioning result based on the trained dual-channel deep learning network. According to the method, collaborative decoupling of the film thickness and the defect signal is achieved through a physical-data fusion method, effective decoupling and independent recognition of the coating film thickness change and the defect response signal are achieved, and the resolution ratio, the accuracy rate and the practicability of a detection system are improved.

Description

Technical Field

[0001] This application relates to the field of microwave nondestructive testing and artificial intelligence fusion technology, specifically to a dual-probe film thickness and defect collaborative decoupling method. Background Technology

[0002] Aero-engine hot-end components often employ multi-layer thermal barrier coatings (Top Coating + Bond Coating) to protect against high-temperature exhaust gas erosion. These coatings operate in harsh environments with intense alternating thermal stresses, making them highly susceptible to failure modes such as thinning, voids, delamination, and interface debonding. Accurately obtaining both "remaining coating thickness" and "hidden defects" parameters simultaneously is crucial for assessing blade remaining life and developing preventative maintenance plans.

[0003] Currently, non-destructive testing methods available for online coating evaluation can be broadly categorized into four types: (1) Ultrasonic / Laser Ultrasonic: Water immersion UT cannot be applied to the whole machine; Dry coupling UT requires coupling agent and the signal is severely attenuated at high temperature, making it difficult to distinguish thickness differences on the order of 100µm and difficult to penetrate ceramic pores to detect micro debonding.

[0004] (2) Electromagnetic eddy current (ECT): Relies on eddy current coupling and is only applicable to conductive substrates and metal coatings; its sensitivity drops sharply when facing low-conductivity ceramic layers, and it cannot provide a reliable assessment of pores or delamination.

[0005] (3) X-ray / CT: High spatial resolution but expensive and has radiation risks; Ni substrate has strong absorption of X-rays, resulting in severe artifacts in CT images and requiring shutdown and disassembly.

[0006] (4) Infrared thermal imaging (IRT): Based on the difference in thermal diffusion, the thickness is quantitatively inverted. However, due to the limitations of heating conditions, thermal property gradients and environmental wind-cooling disturbances, it is difficult to implement in real time on high-temperature service surfaces.

[0007] For microwave detection research, traditional methods often employ a single-probe reflective resonance method. While this method partially verifies the influence of film thickness or debonding on the phase, two core bottlenecks remain unresolved: film thickness and defect signals are severely coupled in the frequency domain, making them indistinguishable with a single probe; and decoupling often relies on empirical thresholds or black-box networks, lacking physical consistency and prone to inaccuracy in complex field noise.

[0008] Therefore, the industry urgently needs an online, interpretable, and physically constrained algorithm framework that can separate signals by leveraging the differentiated frequency sensitivity of dual probes and improve quantitative accuracy by utilizing machine learning, thereby truly solving the engineering challenge of "simultaneous quantitative film thickness measurement and high-resolution defect localization within the same detection window". Summary of the Invention

[0009] To address the aforementioned issues, this application provides a dual-probe film thickness and defect co-decoupling method, aiming to overcome technical bottlenecks such as overlapping spectral responses of film thickness and defect signals and weak generalization of traditional networks.

[0010] The first aspect of this invention provides a method for the coordinated decoupling of film thickness and defects using dual probes, comprising: Microwave reflection signals of the thermal barrier coating are acquired synchronously using a coaxial resonant probe and a CSRR resonant ring probe; the microwave reflection signals include reflection coefficient amplitude, reflection coefficient phase, and resonant frequency; A frequency offset model is constructed based on microwave reflection signals; The dual-channel deep learning network is trained based on the frequency shift model to obtain the trained dual-channel deep learning network. The membrane thickness information and defect localization results are obtained based on the trained dual-channel deep learning network.

[0011] In one alternative implementation, a vector network analyzer is used to excite a coaxial resonant probe and a CSRR resonant ring probe to simultaneously acquire microwave reflection signals from the thermal barrier coating, and wavelet denoising and amplitude-phase normalization are performed on the microwave reflection signals.

[0012] In an optional implementation, the reflection coefficient amplitude and reflection coefficient phase are further weighted:

[0013] in, The weight of the i-th frequency band is... Let be the signal-to-noise ratio of the i-th frequency band, and k be the number of all selected frequency bands.

[0014] In one optional implementation, the frequency offset model constructed based on the frequency offset signal is as follows:

[0015] Where Δf is the resonant frequency shift, Δφ is the phase change of the reflection coefficient, and ΔA is the amplitude change of the reflection coefficient. To assess the system's frequency sensitivity to phase changes, ε represents the system's frequency sensitivity to amplitude variations, and ε is the unavoidable system noise and environmental disturbance term.

[0016] In one alternative implementation, the sensitivity coefficient Obtained through calibration experiments.

[0017] In one optional implementation, the dual-channel deep learning network includes: A dual-input encoder is used to extract the amplitude, phase, and resonant frequency characteristics of the coaxial resonant probe and the CSRR resonant loop probe, respectively. The intermediate layer is used to embed the frequency offset model into the network graph structure; The output layer is used to output film thickness regression values ​​and defect identification probabilities in parallel.

[0018] In one alternative implementation, training the dual-channel deep learning network further includes training based on minimizing the joint loss function: L = Lfilm thickness + Ldefect + λ·Lphysical Where Lfilm thickness is the mean square error of film thickness, Ldefect is the cross-entropy of defects, λ is the weighting coefficient, and Lphysical is the frequency offset residual.

[0019] A second aspect of this invention provides a dual-probe film thickness and defect co-decoupling device, the device comprising: The data acquisition module is used to synchronously acquire microwave reflection signals from the thermal barrier coating using a coaxial resonant probe and a CSRR resonant ring probe; the microwave reflection signals include reflection coefficient amplitude, reflection coefficient phase, and resonant frequency; The model building module is used to build a frequency offset model based on microwave reflection signals; The network training module is used to train a dual-channel deep learning network based on a frequency shift model, resulting in a trained dual-channel deep learning network. The network output module is used to obtain membrane thickness information and defect localization results based on the trained dual-channel deep learning network.

[0020] A third aspect of the present invention provides an electronic device, characterized in that it includes: a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement a dual-probe film thickness and defect co-decoupling method.

[0021] A fourth aspect of the present invention provides a computer-readable storage medium, characterized in that a computer program is stored on the computer-readable storage medium, and the computer program, when executed by a processor, is a method for co-decoupling dual-probe film thickness and defects.

[0022] In this embodiment, the differential responses of coaxial probes and CSRR probes to different physical characteristics are utilized. By embedding a neural network structure with a frequency offset physical model, rapid and accurate separation and simultaneous identification of film thickness and defects are achieved. Through a dual-drive mechanism of "physics-data", a closed loop is completed at three levels: structured mapping of film thickness and defect signals, coupling and decoupling of deep learning networks, and visualization of results. This adapts to the non-destructive testing scenarios of complex layered materials and realizes the accurate, real-time, and highly robust characteristics of microwave non-destructive testing of multilayer heterogeneous materials. Attached Figure Description

[0023] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments of this application will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 This is a flowchart of a dual-probe film thickness and defect co-decoupling method proposed in one embodiment of this application; Figure 2 This is a frequency response offset curve proposed in an embodiment of this application; Figure 3 This is a schematic diagram of the frequency response offset principle proposed in one embodiment of this application; Figure 4 This is a diagram of a collaborative decoupling neural network structure proposed in an embodiment of this application; Figure 5 This is a decoupling output diagram of film thickness and defects proposed in one embodiment of this application; Figure 6 This is a structural diagram of a dual-probe film thickness and defect co-decoupling device proposed in one embodiment of this application; Figure 7 This is a schematic diagram of an electronic device according to this application. Detailed Implementation

[0025] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0026] Please refer to Figure 1 , Figure 1 This is a flowchart of a dual-probe film thickness and defect co-decoupling method proposed in one embodiment of this application. Figure 1 As shown, a dual-probe film thickness and defect co-decoupling method includes: Microwave reflection signals of the thermal barrier coating are acquired synchronously using a coaxial resonant probe and a CSRR resonant ring probe; the microwave reflection signals include reflection coefficient amplitude, reflection coefficient phase, and resonant frequency; A frequency offset model is constructed based on microwave reflection signals; The dual-channel deep learning network is trained based on the frequency shift model to obtain the trained dual-channel deep learning network. The membrane thickness information and defect localization results are obtained based on the trained dual-channel deep learning network.

[0027] In this embodiment, two microwave probes with different frequency response characteristics are used for data acquisition. The coaxial resonant probe exhibits high frequency shift sensitivity to metallic defects (such as cracks and debonding), while the CSRR resonant ring probe shows a highly linear response to changes in the thickness of non-metallic media (such as coating thickness). The acquired microwave reflection signal includes the complex form of the reflection coefficient S11, from which the amplitude (|S11|), phase (∠S11), and resonant frequency are extracted. Features such as...

[0028] A deep learning network architecture with dual input channels is constructed, with each input channel corresponding to data collected by a probe. Amplitude, phase, and frequency features are fed into the encoder, and the physical model is embedded in the intermediate layer in the form of a computational graph. The training objective function simultaneously minimizes the mean square error of the film thickness, the defect cross-entropy, and the physical residual.

[0029] During the inference phase, the system only needs to input the dual-channel S-parameter data of the new sample to simultaneously output the film thickness thermal map and defect point cloud, realizing online, high-precision non-destructive testing. The output is also written to the database for subsequent quality control traceability.

[0030] Please refer to Figure 5 , Figure 5 This is an output diagram of film thickness and defect decoupling proposed in one embodiment of this application. For example... Figure 5 As shown, the algorithm output is presented from the results perspective: the left side is the film thickness thermal distribution of a 10×10 grid in real space, and the right side is the corresponding defect identification map. This figure demonstrates that the present invention can achieve simultaneous output of continuous film thickness regression and point-like defect identification in the spatial domain.

[0031] Furthermore, the microwave reflection signals of the thermal barrier coating are synchronously acquired by exciting the coaxial resonant probe and the CSRR resonant ring probe through a vector network analyzer, and the microwave reflection signals are then subjected to wavelet denoising and amplitude-phase normalization.

[0032] In this embodiment, a vector network analyzer (VNA) is used to excite dual probes to obtain the complex form of the reflection coefficient S11 at multiple frequency points. From the complex form of the reflection coefficient S11, the amplitude (|S11|), phase (∠S11), and resonant frequency are extracted. Features such as wavelet denoising and amplitude-phase normalization are applied to the signal to standardize the input for the model.

[0033] Furthermore, it also includes weighting the amplitude and phase of the reflection coefficient:

[0034] in, The weight of the i-th frequency band is... Let be the signal-to-noise ratio of the i-th frequency band, and k be the number of all selected frequency bands.

[0035] In this embodiment, due to differences in signal quality and probe sensitivity across different frequency bands, a frequency band weighting mechanism is introduced to improve system robustness. Assuming the system operates in N sub-frequency bands, the weight of the i-th frequency band is defined as... ,in This represents the signal-to-noise ratio for that frequency band. The amplitude and phase characteristics of all frequency bands are fed into the network backbone in a weighted manner, making the model pay more attention to the frequency bands with higher information content, suppressing the degradation of predictions by high-noise frequency bands, and improving the anti-interference ability.

[0036] Furthermore, the frequency offset model constructed based on the frequency offset signal is as follows:

[0037] Where Δf is the resonant frequency shift, Δφ is the phase change of the reflection coefficient, and ΔA is the amplitude change of the reflection coefficient. To assess the system's frequency sensitivity to phase changes, ε represents the system's frequency sensitivity to amplitude variations, and ε is the unavoidable system noise and environmental disturbance term.

[0038] In this embodiment, changes in the target's physical properties (film thickness or defects) during the probe response lead to changes in the equivalent dielectric constant along the electromagnetic propagation path, resulting in a shift in the resonant frequency. The total frequency shift Δf can be defined as a linear combination of the phase change Δφ and the amplitude change ΔA. This model serves as both the probe group calibration equation and a hard constraint for subsequent networks. This model reveals the mapping relationship between electromagnetic characteristic parameters and structural physical changes, providing the fundamental physical basis for the decoupling process.

[0039] Please refer to Figure 2 , Figure 2 This is a frequency response offset curve proposed in one embodiment of this application. As shown in Figure 2, the curve gives the resonant response before the change ( ")" and "after the change ()" Two envelope curves, '′)”. The rightward shift of the peak value of the curve envelope corresponds to Δf, and the difference in peak value reflects the different attenuation characteristics of the resonance intensity under the influence of defects and film thickness variations by the dual probes. This graphical result further verifies that… Figure 2 The observability and measurability of the Δf linear model.

[0040] Please refer to Figure 3 , Figure 3 This is a schematic diagram of the frequency response offset principle proposed in one embodiment of this application. For example... Figure 3As shown, the electromagnetic link between film thickness variation Δd and defect disturbance δ via "equivalent dielectric constant change Δε_eff → phase change Δφ / amplitude change ΔA → frequency shift Δf" is visually illustrated in the figure. The formulas given in the figure... The one-to-one correspondence between the electromagnetic features and physical geometric quantities of the dual-probe model was clarified, laying the foundation for the physical prior embedding of the subsequent network model.

[0041] Furthermore, the sensitivity coefficient Obtained through calibration experiments.

[0042] Furthermore, the dual-channel deep learning network includes: A dual-input encoder is used to extract the amplitude, phase, and resonant frequency characteristics of the coaxial resonant probe and the CSRR resonant loop probe, respectively. The intermediate layer is used to embed the frequency offset model into the network graph structure; The output layer is used to output film thickness regression values ​​and defect identification probabilities in parallel.

[0043] In this embodiment, the network consists of the following key modules: Dual-input encoder: processes input tensors from the coaxial probe and CSRR probe respectively, and extracts their temporal features such as amplitude, phase, and frequency; Intermediate layer: The physical model from the previous step is embedded into the network graph structure as an intermediate constraint term during network training. This guides the deep learning network to minimize both prediction error and physical offset residuals during optimization. The joint loss function is defined as: L = Lfilm thickness + Ldefect + λ·Lphysical, where... ; Output layer: Branch output 1 is the film thickness regression value (continuous variable, unit μm), and output 2 is the defect identification probability (0 / 1 classification problem). This structure realizes automatic decoupling of structural signals, channel feature sharing, and enhanced physical consistency of the model.

[0044] Please refer to Figure 4 , Figure 4 This is a diagram of a collaborative decoupling neural network structure proposed in an embodiment of this application. For example... Figure 4 As shown, L_thickness is the film thickness error, L_defect is the defect classification cross-entropy, and L_phy is the frequency offset residual. The core network topology of the algorithm is described: the dual-channel input is mapped to two encoder branches respectively, and then embedded in the "Physics-EmbeddingLayer". Figure 1The Δf physical model is used; downstream, a fusion layer and dual output heads are set to regress the film thickness value and classify the defect state, respectively. In order to ensure that the network training and physical mechanism are coordinated, the loss function is L=L_thickness+L_defect+λ·L_phy, where L_phy is the physical offset residual term, to ensure that the network prediction is consistent with the Δf formula.

[0045] Furthermore, training a dual-channel deep learning network also includes training based on minimizing the joint loss function: L = Lfilm thickness + Ldefect + λ·Lphysical Where Lfilm thickness is the mean square error of film thickness, Ldefect is the cross-entropy of defects, λ is the weighting coefficient, and Lphysical is the frequency offset residual.

[0046] In this embodiment, during the training phase, a large number of labeled samples with different film thicknesses and defect states are used, and the Adam optimizer is used to backpropagate the joint loss function to adjust the network parameters to the optimal level.

[0047] Please refer to Figure 6 , Figure 6 This is a structural diagram of a dual-probe film thickness and defect co-decoupling device proposed in one embodiment of this application. Figure 6 As shown in the figure, this disclosure also provides a dual-probe film thickness and defect co-decoupling device, the device comprising: a data acquisition module 601, a model building module 602, a network training module 603, and a network output module 604; wherein, The data acquisition module 601 is used to synchronously acquire microwave reflection signals of the thermal barrier coating through a coaxial resonant probe and a CSRR resonant ring probe; the microwave reflection signals include reflection coefficient amplitude, reflection coefficient phase and resonant frequency; Model building module 602 is used to build a frequency offset model based on microwave reflection signals; The network training module 603 is used to train a dual-channel deep learning network based on a frequency shift model to obtain the trained dual-channel deep learning network. The network output module 604 is used to obtain membrane thickness information and defect localization results based on the trained dual-channel deep learning network.

[0048] This disclosure also provides an electronic device, please refer to... Figure 7 , Figure 7 This is a schematic diagram of an electronic device illustrated in an embodiment of this disclosure. For example... Figure 7As shown, the electronic device 100 includes a memory 110 and a processor 120. The memory 110 and the processor 120 are connected via a bus for communication. The memory 110 stores a computer program that can run on the processor 120 to implement the steps in the dual-probe film thickness and defect co-decoupling method disclosed in this embodiment.

[0049] The disclosed embodiments also provide a computer-readable storage medium that, when the instructions in the computer-readable storage medium are executed by a processor of a computer device, enables the computer device to perform the steps in the dual-probe film thickness and defect co-decoupling method of the present disclosure embodiments.

[0050] In summary, this invention proposes for the first time a decoupling mechanism based on the frequency response difference of dual probes, effectively distinguishing the electromagnetic responses of two physical characteristics: film thickness and defects. Furthermore, it introduces frequency offset physical constraints into the deep learning network, improving decoupling accuracy and the model's physical interpretation capability. This method integrates the advantages of physical modeling and deep learning, making it suitable for non-destructive testing scenarios involving complex layered materials. Experiments show that the algorithm has a film thickness prediction error of less than ±5μm and an accuracy rate exceeding 95% for identifying defects at the 0.5mm level, demonstrating superior stability and robustness compared to traditional neural network models.

[0051] Simultaneous Dual Measurement: Relying on the differentiated resonance characteristics of dual probes and physical constraint neural networks, this invention can simultaneously output 20µm-level film thickness distribution and 0.5mm-level defect location results in a single scan, truly realizing integrated online detection of "thickness + defect".

[0052] Real-time and efficient: inference latency is less than 50ms, and it can be embedded in FPGA / ARM industrial control boards; the whole machine volume is less than 1L, and it can be seamlessly integrated into existing VNA production lines or maintenance terminals to achieve 24-hour continuous monitoring.

[0053] Physically explainable: through hard constraints The residual loss is minimized, and the network output always conforms to the electromagnetic field mechanism, avoiding distortion of the black box model when transferring to on-site working conditions.

[0054] Noise robustness: Multi-band SNR-weighted suppression of high-noise subbands; dual-probe complementary fault tolerance, single-path degradation does not lead to overall failure, and prediction accuracy remains >90% even when the ambient noise increases by 15dB.

[0055] Safe and economical: No ionizing radiation, power consumption <5W, cost reduction of 80% compared to CT / RT, suitable for high-temperature enclosed or radiation-sensitive scenarios.

[0056] The platform is scalable: the algorithm framework can be adapted to other microwave probe arrays or multiphysics (temperature / stress) coupling tasks, providing a unified data interface for the aerospace, energy and composite materials industries.

[0057] This application describes embodiments with reference to flowchart illustrations and / or block diagrams of methods, apparatuses, electronic devices, and computer program products according to embodiments of this application. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal device, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0058] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing terminal device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0059] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal equipment, causing a series of operational steps to be performed on the computer or other programmable terminal equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable terminal equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0060] Although preferred embodiments of the present application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the embodiments of the present application.

[0061] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0062] Although preferred embodiments of the present application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the embodiments of the present application.

[0063] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes said element.

[0064] The above provides a detailed description of the dual-probe film thickness and defect co-decoupling method provided in this application. Specific examples have been used to illustrate the principle and implementation of this application. The description of the above embodiments is only for the purpose of helping to understand the method and core idea of ​​this application. At the same time, for those skilled in the art, there will be changes in the specific implementation and application scope based on the idea of ​​this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for co-decoupling film thickness and defects using dual probes, characterized in that, include: Microwave reflection signals of the thermal barrier coating are acquired synchronously using a coaxial resonant probe and a CSRR resonant ring probe; the microwave reflection signals include reflection coefficient amplitude, reflection coefficient phase, and resonant frequency; A frequency offset model is constructed based on microwave reflection signals; The dual-channel deep learning network is trained based on the frequency shift model to obtain the trained dual-channel deep learning network. The membrane thickness information and defect localization results are obtained based on the trained dual-channel deep learning network.

2. The dual-probe film thickness and defect co-decoupling method according to claim 1, characterized in that, The microwave reflection signal of the thermal barrier coating is synchronously acquired by exciting the coaxial resonant probe and the CSRR resonant ring probe using a vector network analyzer, and then the microwave reflection signal is processed by wavelet denoising and amplitude-phase normalization.

3. The dual-probe film thickness and defect co-decoupling method according to claim 1, characterized in that, This also includes weighting the amplitude and phase of the reflection coefficient: in, The weight of the i-th frequency band is... Let be the signal-to-noise ratio of the i-th frequency band, and k be the number of all selected frequency bands.

4. The dual-probe film thickness and defect co-decoupling method according to claim 1, characterized in that, The frequency offset model constructed based on the frequency offset signal is as follows: Where Δf is the resonant frequency shift, Δφ is the phase change of the reflection coefficient, and ΔA is the amplitude change of the reflection coefficient. To assess the system's frequency sensitivity to phase changes, ε represents the system's frequency sensitivity to amplitude variations, and ε is the unavoidable system noise and environmental disturbance term.

5. The dual-probe film thickness and defect co-decoupling method according to claim 4, characterized in that, The sensitivity coefficient Obtained through calibration experiments.

6. The dual-probe film thickness and defect co-decoupling method according to claim 1, characterized in that, The dual-channel deep learning network includes: A dual-input encoder is used to extract the amplitude, phase, and resonant frequency characteristics of the coaxial resonant probe and the CSRR resonant loop probe, respectively. The intermediate layer is used to embed the frequency offset model into the network graph structure; The output layer is used to output film thickness regression values ​​and defect identification probabilities in parallel.

7. The dual-probe film thickness and defect co-decoupling method according to claim 6, characterized in that, Training a dual-channel deep learning network also includes training by minimizing the joint loss function: L = Lfilm thickness + Ldefect + λ·Lphysical Where Lfilm thickness is the mean square error of film thickness, Ldefect is the cross-entropy of defects, λ is the weighting coefficient, and Lphysical is the frequency offset residual.

8. A dual-probe film thickness and defect co-decoupling device according to any one of claims 1-7, characterized in that, The device includes: The data acquisition module is used to synchronously acquire microwave reflection signals from the thermal barrier coating using a coaxial resonant probe and a CSRR resonant ring probe; the microwave reflection signals include reflection coefficient amplitude, reflection coefficient phase, and resonant frequency; The model building module is used to build a frequency offset model based on microwave reflection signals. The network training module is used to train a dual-channel deep learning network based on a frequency shift model, resulting in a trained dual-channel deep learning network. The network output module is used to obtain membrane thickness information and defect localization results based on the trained dual-channel deep learning network.

9. An electronic device, characterized in that, include: A processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements the dual-probe film thickness and defect co-decoupling method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, A computer program is stored on the computer-readable storage medium, which, when executed by a processor, implements the dual-probe film thickness and defect co-decoupling method as described in any one of claims 1 to 7.

Images