Electrolytic capacitor testing method and system

Abstract

This invention discloses a method and system for testing electrolytic capacitors, specifically relating to the field of electrolytic capacitor testing technology. It acquires time-frequency coupled waveforms for electrolytic capacitor testing, constructs an anti-interference benchmark feature library covering the process fluctuation range based on the correlation function between the electrolyte dynamic impedance spectrum and ultrasonic velocity, dynamically generates a graph neural network topology update rule based on the physical equations of electrolyte evaporation and electrode corrosion, and a physically constrained graph topology structure. It extracts defect-sensitive frequency band features of cross-modal signals through a differentiable frequency band selection network, extracts multi-scale fusion features from the time-frequency coupled waveform based on these defect-sensitive frequency band features, inputs the multi-scale fusion features into an edge-preserving regularization layer, and adaptively adjusts the reconstruction parameters according to the confidence level of the defect gradient field to generate a three-dimensional defect morphology that satisfies continuity constraints, thus solving the problem of low detection accuracy of internal defects in electrolytic capacitors.

Description

Technical Field

[0001] This invention relates to the field of electrolytic capacitor testing technology, and more specifically, to an electrolytic capacitor testing method and system. Background Technology

[0002] Electrolytic capacitors, widely used energy storage components in electronic devices, are directly threatened by internal defects (such as electrolyte leakage and plate corrosion). Traditional testing methods rely on destructive disassembly or sampling inspection (such as dissection analysis and X-ray imaging), which suffer from low efficiency, insufficient coverage, and high cost. Existing non-destructive testing technologies (such as single-modal ultrasonic scanning or conductivity measurement) are limited by signal sensitivity and feature decoupling capabilities, making it difficult to accurately identify micron-level defects and unable to achieve three-dimensional defect localization, resulting in significant blind spots in quality control.

[0003] In existing technologies, the non-destructive testing of internal defects in electrolytic capacitors faces the following core challenges:

[0004] The mapping relationship between indirect signals and internal defects is complex, and traditional single-mode detection methods struggle to extract highly discriminative features, resulting in a high false positive rate.

[0005] The lack of an effective fusion mechanism for multi-source heterogeneous data (such as acoustic and electromagnetic signals) restricts the accuracy of defect classification and location.

[0006] There is an urgent need for an efficient, high-precision, and cost-controllable non-destructive testing method to overcome the bottleneck of analyzing the concealment of complex defects and to optimize the testing of electrolytic capacitors. Summary of the Invention

[0007] To overcome the aforementioned deficiencies of the prior art, the present invention provides an electrolytic capacitor testing method and system, which solves the problems mentioned in the background art by synchronously acquiring multimodal physical signals, dynamically compensating and modeling physical constraints.

[0008] To achieve the objective of this invention, a method for testing electrolytic capacitors is provided, comprising the following steps:

[0009] Step 1: Synchronous acquisition of multimodal physical coupling signals: Establish a phase-locked trigger acquisition mechanism for ultrasonic and electromagnetic waves, and generate time-frequency coupled waveforms of dual-mode signals through bus-level hardware synchronous control; Based on the correlation function Φ(f) between the dynamic impedance spectrum of the electrolyte and the ultrasonic velocity, construct an anti-interference reference feature library covering the process fluctuation range, and dynamically compensate for the ultrasonic propagation speed according to the ratio of the physical property parameters of the current electrolyte and the reference electrolyte;

[0010] Step 2: Cross-modal modeling of physical constraints: Based on the physical equations of electrolyte evaporation and electrode corrosion, the topology update rules of the graph neural network are dynamically generated, and the graph topology of physical constraints is output. The defect-sensitive frequency band features of the cross-modal signal are extracted through a differentiable frequency band selection network, and dynamic attention weights are generated. The graph topology of physical constraints is used to define the spatial correlation of the defect diffusion path.

[0011] Step 3: Edge-Adaptive Defect Morphology Inversion: Based on the defect-sensitive frequency band characteristics, multi-scale fusion features are extracted from the time-frequency domain coupled waveform. The multi-scale fusion features are input into the edge-preserving regularization layer. The reconstruction parameters are adaptively adjusted according to the confidence level of the defect gradient field to generate a three-dimensional defect morphology that satisfies the continuity constraint. The physical consistency of the inversion results is optimized based on the difference in propagation characteristics between ultrasonic waves and electromagnetic waves. (The physical consistency of the optimized inversion results is verified by judging the physical laws of the ultrasonic wave propagation time difference and the electromagnetic wave phase difference. The three-dimensional defect morphology refers to the voxel grid of internal defects of the electrolytic capacitor generated by the inversion network. Each voxel value represents the probability of the presence of a defect at the corresponding location.)

[0012] Preferably, by traversing different gradient electrolyte formulations, encapsulation stresses, and temperature combinations, benchmark feature vectors are collected. Each set of benchmark feature vectors includes the corrected ultrasonic velocity, the correlation function Φ(f) between the electrolyte dynamic impedance spectrum and the ultrasonic velocity attenuation, the extracted frequency domain amplitude spectrum of the ultrasonic amplitude sequence after Fourier transform, and the frequency domain gradient of the electromagnetic wave impedance spectrum. The benchmark feature vectors are then summarized to obtain an anti-interference benchmark feature library; based on the electrolyte dynamic impedance spectrum Z... electrolyte (f) with ultrasonic velocity v g The correlation function Φ(f) is obtained by calculating the correlation function Φ(f) using the following formula, where ω represents the angular frequency and f represents the frequency. This provides a physical basis for cross-modal modeling:

[0013]

[0014] By dynamically compensating for electrolyte viscosity and dielectric constant, the influence of process fluctuations on ultrasonic velocity is eliminated; the current electrolyte viscosity and dielectric constant are obtained and denoted as η. c ε c The viscosity and dielectric constant of the reference electrolyte are denoted as η. b ε b The dynamic compensation formula automatically corrects for the impact of process fluctuations (such as batch differences in electrolyte) on the ultrasonic propagation speed, ensuring consistent testing conditions. The dynamic compensation formula for ultrasonic speed is as follows:

[0015]

[0016] v comp This indicates the corrected ultrasonic velocity.

[0017] Preferably, based on an anti-interference benchmark feature library, the physical laws of electrolyte evaporation and plate corrosion are embedded into the topology update and feature fusion process of a graph neural network to solve the misjudgment problem caused by the reliance on statistical correlation in traditional multimodal fusion.

[0018] The execution process of the topology update rule in a graph neural network includes:

[0019] The electrolyte evaporation process is modeled as a concentration diffusion equation, and the electrode corrosion process is modeled as an electrochemical reaction equation.

[0020] The physical equations for electrolyte evaporation and electrode corrosion include concentration diffusion equations and electrochemical reaction equations;

[0021] The concentration diffusion equation and the electrochemical reaction equation are combined and discretized into the ratio of the concentration gradient difference between nodes to the compensation rate. This ratio is then multiplied by the correlation function to generate the edge weight calculation rule, where the edge weight w is... ij It is obtained by multiplying the absolute value of the electrolyte concentration gradient difference to the compensated ultrasonic group velocity by the correlation function Φ(f);

[0022] Adjust the edge weights according to the actual process parameters to ensure that the message transmission path of the graph neural network conforms to the physical laws of material corrosion diffusion;

[0023] The implementation method of the differentiable frequency band selection network is as follows:

[0024] Input signal: Frequency domain characteristics of the ultrasonic time-domain signal and electromagnetic wave complex impedance spectrum acquired in step one;

[0025] Defect-sensitive frequency band extraction: Calculate the product of the absolute value of the frequency domain gradient of the ultrasonic amplitude spectrum and the absolute value of the frequency domain gradient of the electromagnetic impedance spectrum to generate the energy distribution ΔE(f) of the defect-sensitive frequency band;

[0026] Dynamic attention weight generation: The energy distribution of defect-sensitive frequency bands is normalized, and a frequency band attention mask is generated by filtering through a preset threshold to suppress noise interference in non-defect frequency bands;

[0027] The multi-scale feature fusion and node update method is as follows:

[0028] Input features: The baseline feature vector constructed in step one is used as the initial node features of the graph neural network;

[0029] Feature propagation rule: The feature update of node i depends on the weighted sum of features of its neighboring nodes j, with the weights being the edge weights w. ij The product of the frequency band value corresponding to the frequency band attention mask;

[0030] Output data: The updated node features are passed to step three for defect morphology inversion.

[0031] Preferably, the topology update rule includes an edge weight update formula, which is obtained as follows:

[0032] Let nodes i and j represent the indices of the feature vectors of different spatial locations inside the electrolytic capacitor; the edge weights are obtained using the following formula;

[0033]

[0034] Where, ΔC ij This represents the difference in electrolyte concentration gradient between nodes i and j;

[0035] The edge weights are summed to obtain the edge weight matrix.

[0036] Preferably, the method for obtaining the multi-scale fusion features is as follows:

[0037] Wavelet transform is performed on ultrasonic time-domain signals to extract multi-resolution features (macroscale is used to capture overall corrosion, and microscale is used to locate micro-defects), and short-time Fourier transform is performed on electromagnetic wave complex impedance spectrum to obtain local frequency domain energy distribution.

[0038] Based on the frequency band attention mask, the ultrasonic wavelet coefficients and electromagnetic wave frequency domain energy are weighted and fused to enhance the signal strength in the defect-sensitive frequency band.

[0039] By using the message passing mechanism of graph neural networks, the weighted cross-modal features are aggregated in layers according to different scales (macro, meso, and micro). Combined with the topological constraints of the electrolyte diffusion path, spatial-frequency-temporal three-dimensional fusion features are generated, providing highly discriminative input for defect inversion.

[0040] Preferably, the process of obtaining the three-dimensional defect morphology includes:

[0041] Input multi-scale fusion features, calculate the defect gradient field of the defect boundary through the spatial gradient of the multi-scale fusion features, and quantify the confidence of the defect boundary;

[0042] The encoder-decoder architecture is used to obtain the voxel mesh corresponding to the 3D defect morphology. The encoder compresses the fused features into a low-dimensional latent vector through a 3D convolutional layer to extract the macroscopic defect distribution. The decoder gradually upsamples through a 3D deconvolutional layer and updates the voxel mesh by combining the physical constraint graph topology. The adjacency matrix of the physical constraint graph topology serves as the spatial connection template for the decoder, which restricts the diffusion path of the defect morphology during the upsampling process to be consistent with the actual physical law of electrolyte corrosion. This forces a smooth transition of defect probabilities between adjacent spatial locations to avoid isolated noise points.

[0043] The ultrasonic wave propagation time difference and electromagnetic wave phase difference are calculated, and the calculation results drive the correction of the three-dimensional defect morphology. If the ultrasonic wave propagation time difference exceeds the threshold, it indicates that the material density in the defect area is reduced or there are voids. The voxel value (defect probability) at the corresponding position is increased. If the electromagnetic wave phase difference is lower than the requirement, it indicates that the conductivity of the electrolyte is reduced or the electrode is corroded. The boundary confidence of the corrosion area is enhanced.

[0044] Preferably, the regularization coefficient is dynamically adjusted based on the confidence level of the defect gradient field, the non-equilibrium entropy yield, or a fusion of the two, and the regularization coefficient is used to control the smoothing intensity.

[0045] Preferably, the regularization coefficient is dynamically adjusted based on the confidence level of the defect gradient field:

[0046] For high-confidence regions (defect gradient field confidence ≥ confidence threshold), reduce the regularization coefficient and prioritize preserving the sharp features of the defect boundary;

[0047] For low-confidence regions (defect gradient field confidence < confidence threshold), enhance the regularization coefficient to suppress artifact interference caused by noise;

[0048] The confidence threshold is adaptively set based on the gradient field statistical distribution of historical defect samples; after each batch of testing is completed, the threshold is dynamically updated based on the confidence distribution of the defect gradient field of the current sample.

[0049] Preferably, the regularization coefficient is dynamically adjusted based on the non-equilibrium entropy yield.

[0050] Based on the electrolyte temperature field T(x, y, z) and the dissipation function σ(x, y, z), the formula is used to... Calculate the non-equilibrium entropy yield of the defect region. Non-equilibrium entropy yield represents the amount of energy dissipated in the defect region per unit time;

[0051] The electrolyte temperature field is measured by an infrared thermal imager or temperature sensor; the dissipation function is used to characterize the energy dissipation density of the material corrosion process; Ω represents the spatial extent of the defect region.

[0052] If the non-equilibrium entropy yield threshold is exceeded Increase the regularization coefficient to enhance smoothing strength and suppress noise interference in high dissipation regions;

[0053] If the non-equilibrium entropy yield threshold is not exceeded, the regularization coefficient is reduced to preserve defect details in the low-dissipation region.

[0054] Preferably, the fusion dynamic adjustment regularization coefficient based on the defect gradient field confidence and the non-equilibrium entropy yield includes:

[0055] By combining the confidence level of the signal defect gradient field with the non-equilibrium entropy yield, the limitations of a single dimension can be overcome.

[0056] Let the regularization coefficient be denoted as λ, λ0 represent the initial regularization coefficient, and T be the confidence level of the defect gradient field. z ,

[0057]

[0058] Among them, tanh(·) is used to compress the entropy production ratio to the (-1, 1) interval to avoid extreme values ​​causing the regularization coefficient to run out of control.

[0059] To achieve the objective of this invention, the present invention provides an electrolytic capacitor testing system, comprising:

[0060] The test signal synchronous acquisition module synchronously acquires ultrasonic and electromagnetic wave signals through a phase-locked trigger acquisition mechanism, generates a time-frequency domain coupled waveform data stream, and transmits it to the dynamic compensation module;

[0061] The dynamic compensation module constructs an anti-interference benchmark feature library covering the process fluctuation range. Based on the ratio of the current physical property parameters to the benchmark electrolyte, it dynamically compensates for the ultrasonic propagation speed, calculates the corrected ultrasonic speed, and transmits it to the graph engine module; it also transmits the anti-interference benchmark feature library to the frequency selection network module.

[0062] The graph engine module dynamically generates the topology update rules of the graph neural network based on the physical equations of electrolyte evaporation and electrode corrosion, and outputs the graph topology structure with physical constraints.

[0063] The frequency selection network module is used to calculate the energy distribution of the defect-sensitive frequency band by multiplying the gradient across the modal frequency domain. After normalization, it generates a frequency band attention mask and transmits it to the fusion engine module.

[0064] The fusion engine module extracts multi-scale fusion features from time-frequency coupled waveforms based on defect-sensitive frequency band features, inputs the multi-scale fusion features into the edge-preserving regularization layer, adaptively adjusts the reconstruction parameters according to the confidence of the defect gradient field, generates a three-dimensional defect morphology that meets the continuity constraint, and transmits it to the inversion optimization module.

[0065] The inversion optimization module generates a voxel mesh through an encoder-decoder architecture and applies an edge-preserving regularization layer. It obtains a counterfactual inference loss function based on the weighted summation of the ultrasonic wave propagation time difference and the electromagnetic wave phase difference, and transmits it to the graph engine module to update the weights of the 3D convolution kernel, thereby achieving closed-loop feedback.

[0066] The technical effects and advantages of this invention are as follows:

[0067] (1) The invention provides a method for testing electrolytic capacitors, establishes a phase-locked trigger synchronous acquisition mechanism for ultrasonic and electromagnetic waves, constructs an anti-interference benchmark feature library by combining the correlation function between the dynamic impedance spectrum of the electrolyte and the ultrasonic velocity, and corrects the influence of process fluctuations on the ultrasonic propagation speed through a dynamic compensation formula to ensure the authenticity and consistency of multimodal signals; based on the physical equations of electrolyte evaporation and plate corrosion, the material corrosion diffusion law is embedded into the edge weight calculation rules of the graph neural network to constrain the message transmission path to be consistent with the actual physical process, thereby avoiding the misjudgment problem of traditional multimodal fusion relying on statistical correlation.

[0068] (2) The invention provides a method for testing electrolytic capacitors, which generates a three-dimensional defect morphology that satisfies continuity constraints based on multi-scale fusion features and edge-preserving regularization technology; extracts macro-micro fusion features by wavelet transform of ultrasonic time-domain signals and short-time Fourier transform of electromagnetic frequency-domain features, and enhances defect-sensitive frequency bands by combining frequency band attention masking; reconstructs a three-dimensional voxel grid using an encoder-decoder architecture to force a smooth transition of defect probabilities to avoid isolated noise points; introduces a physical consistency verification mechanism between ultrasonic propagation time difference and electromagnetic phase difference; and balances defect detail preservation and noise suppression by dynamically adjusting the regularization coefficient, ultimately achieving high-fidelity inversion of internal defects of electrolytic capacitors, meeting the accuracy and reliability requirements of industrial testing. Attached Figure Description

[0069] Figure 1 This is a flowchart of the electrolytic capacitor testing method of the present invention.

[0070] Figure 2 This is a block diagram of the electrolytic capacitor testing system of the present invention. Detailed Implementation

[0071] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0072] At the same time, it should be understood that, for ease of description, the dimensions of the various parts shown in the accompanying drawings are not drawn according to actual scale.

[0073] The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit the scope of this application and its application or use.

[0074] Techniques, methods, and equipment known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and equipment should be considered part of the specification.

[0075] Existing non-destructive testing techniques for internal defects in electrolytic capacitors have the following problems:

[0076] 1. Single-modal detection methods struggle to extract highly discriminative features, leading to a high false positive rate. This is because the mapping relationship between indirect signals and internal defects is complex, and single-modal detection methods cannot obtain enough information to accurately identify defects.

[0077] 2. The lack of an effective fusion mechanism for multi-source heterogeneous data (such as acoustic and electromagnetic signals) limits the accuracy of defect classification and location. This is because there is no unified processing and analysis method for different types of signal data, making it impossible to fully utilize the information contained in multiple signals. Based on this, the present invention provides the technical solution of Embodiment 1.

[0078] Example 1, see Figure 1 The present invention provides a flowchart of a method for testing electrolytic capacitors. Figure 1 The electrolytic capacitor testing method shown includes the following steps:

[0079] Step 1: Synchronous acquisition of multimodal physical coupling signals: Establish a phase-locked trigger acquisition mechanism for ultrasonic and electromagnetic waves, and generate time-frequency coupled waveforms of dual-mode signals through bus-level hardware synchronous control; Based on the correlation function Φ(f) between the dynamic impedance spectrum of the electrolyte and the ultrasonic velocity, construct an anti-interference reference feature library covering the process fluctuation range, and dynamically compensate for the ultrasonic propagation velocity according to the ratio of the physical property parameters of the current electrolyte and the reference electrolyte;

[0080] The explanation is that by constructing a benchmark feature library resistant to process interference, signal changes caused by actual defects can be identified more accurately without being affected by normal fluctuations in the production process. This helps reduce false alarm rates and improves the accuracy and reliability of detection. For example, in the production of electrolytic capacitors, the composition and concentration of the electrolyte may have slight batch-to-batch variations, which can affect the propagation characteristics of ultrasound inside the capacitor. By establishing a benchmark feature library containing ultrasound propagation parameters under different electrolyte formulations and concentrations, the detection parameters can be dynamically adjusted according to the characteristics of the electrolyte in the current batch. If the viscosity of a certain batch of electrolyte is slightly higher, causing the ultrasound propagation speed to slow down, this change can be automatically compensated for based on the data in the benchmark feature library, ensuring the consistency and accuracy of the detection results.

[0081] Step 2: Cross-modal modeling of physical constraints: Based on the physical equations of electrolyte evaporation and electrode corrosion, the topology update rules of the graph neural network are dynamically generated, and the graph topology of physical constraints is output. The defect-sensitive frequency band features of the cross-modal signal are extracted through a differentiable frequency band selection network, and dynamic attention weights are generated. The graph topology of physical constraints is used to define the spatial correlation of the defect diffusion path.

[0082] The explanation explains that by automatically selecting defect-sensitive frequency band features, signal changes related to defects are captured, reducing interference from irrelevant information. Dynamic attention weighting adaptively adjusts the importance of each modal signal according to different defect types and detection conditions, thereby improving detection flexibility and accuracy. For example, when detecting electrolyte leakage in electrolytic capacitors, it is found that ultrasonic signals are most sensitive to leakage in the 20-30kHz band, while electromagnetic signals react most significantly in the 1-2MHz band. The differentiable band selection network automatically highlights the characteristics of these frequency bands while suppressing noise in other bands. Furthermore, if electrolyte leakage is detected as primarily affecting ultrasonic signals, dynamically increasing the weight of ultrasonic modes and decreasing the weight of electromagnetic modes ensures that the system relies more on the most relevant and reliable information to make judgments.

[0083] Step 3: Edge-Adaptive Defect Morphology Inversion: Based on the defect-sensitive frequency band characteristics, multi-scale fusion features are extracted from the time-frequency domain coupled waveform. These multi-scale fusion features are input into the edge-preserving regularization layer. The reconstruction parameters are adaptively adjusted according to the defect gradient field confidence level to generate a three-dimensional defect morphology that satisfies continuity constraints. Based on the difference in propagation characteristics between ultrasonic waves and electromagnetic waves, the physical consistency of the inversion results is optimized.

[0084] The explanation explains that by judging the physical laws of the time difference of ultrasonic wave propagation and the phase difference of electromagnetic wave, the physical consistency of the optimized inversion results is verified; the three-dimensional defect morphology refers to the voxel grid of internal defects of the electrolytic capacitor generated by the inversion network, and each voxel value represents the probability of the existence of a defect at the corresponding location.

[0085] The explanation is that optimizing the physical consistency of inversion results based on the differences in propagation characteristics between ultrasound and electromagnetic waves refers to better interpreting and fusing information from different modes by considering the differences in the propagation characteristics of ultrasound and electromagnetic waves inside electrolytic capacitors. This reduces errors caused by a single mode and improves the accuracy and reliability of the reconstruction results, especially when dealing with complex or multiple defects. For example, when detecting corrosion on the plates of an electrolytic capacitor, ultrasound is better at detecting physical structural changes caused by corrosion, while electromagnetic waves are more sensitive to changes in electrical properties caused by corrosion. Suppose ultrasound data shows a 5mm x 5mm structural anomaly in a certain area, while electromagnetic wave data indicates that the electrical anomaly range in that area is 6mm x 6mm. Considering both types of information, it is possible to conclude that the core corrosion area is 5mm x 5mm, but its influence extends to 6mm x 6mm. This fusion method considers both physical structural changes and electrical property changes, thus obtaining a more comprehensive and accurate defect description.

[0086] Furthermore, to address the impact of strong electromagnetic interference in the production line environment on electromagnetic wave signal acquisition, an adaptive electromagnetic shielding and signal enhancement method is proposed, including the following steps:

[0087] Dynamic electromagnetic shielding: An adjustable multi-layer electromagnetic shielding structure is designed around the electromagnetic wave sensor. The multi-layer electromagnetic shielding structure consists of multiple independently controlled metal mesh layers. The aperture and conductivity of each mesh layer are adjusted in real time through microelectromechanical systems (MEMS) technology.

[0088] Real-time monitoring of electromagnetic interference intensity and spectrum distribution in the environment.

[0089] Adaptive shielding control: Based on environmental interference monitoring results, the parameters of the multi-layer shielding structure are adjusted in real time using a fuzzy control algorithm; for example, when strong interference in a certain frequency band is detected, the mesh aperture and conductivity of a specific layer are adjusted accordingly to suppress the interference in that frequency band to the greatest extent.

[0090] Signal enhancement processing: Adaptive filtering and enhancement processing are performed on the acquired electromagnetic wave signals; wavelet transform is used to decompose the signals into different frequency bands, and Wiener filters are applied to the signals of each frequency band according to the environmental interference monitoring results to suppress noise;

[0091] Dynamic parameter optimization: A feedback loop is established to continuously optimize the shielding structure parameters and signal processing algorithm parameters based on signal quality indicators (such as signal-to-noise ratio). A genetic algorithm is used to achieve dynamic optimization of the multi-layer electromagnetic shielding structure parameters to adapt to different interference environments.

[0092] The explanation is that by combining dynamically adjusted physical shielding with intelligent signal processing, electromagnetic interference in complex production environments is suppressed, significantly improving the quality of electromagnetic wave signal acquisition. In this way, the quality of electromagnetic wave signal acquisition is guaranteed even in production line environments with strong electromagnetic interference, thereby improving the accuracy and reliability of defect detection in electrolytic capacitors.

[0093] In this embodiment of the invention, it is necessary to further explain that by traversing different gradient electrolyte formulations, encapsulation stresses, and temperature combinations, benchmark feature vectors are collected. Each set of benchmark feature vectors includes the corrected ultrasonic velocity, the correlation function Φ(f) between the electrolyte dynamic impedance spectrum and the ultrasonic velocity attenuation, the extracted frequency domain amplitude spectrum of the ultrasonic amplitude sequence after Fourier transform, and the frequency domain gradient of the electromagnetic wave impedance spectrum. The benchmark feature vectors are summarized to obtain an anti-interference benchmark feature library; based on the electrolyte dynamic impedance spectrum Z... electrolyte (f) with ultrasonic velocity v g The correlation function Φ(f) is obtained by calculating the correlation function Φ(f) using the following formula, where ω represents the angular frequency and f represents the frequency. This provides a physical basis for cross-modal modeling:

[0094]

[0095] By dynamically compensating for electrolyte viscosity and dielectric constant, the influence of process fluctuations on ultrasonic velocity is eliminated; the current electrolyte viscosity and dielectric constant are obtained and denoted as η. c ε c The viscosity and dielectric constant of the reference electrolyte are denoted as η. b ε b The dynamic compensation formula automatically corrects for the impact of process fluctuations (such as batch differences in electrolyte) on the ultrasonic propagation speed, ensuring consistent testing conditions. The dynamic compensation formula for ultrasonic speed is as follows:

[0096]

[0097] v comp This indicates the corrected ultrasonic velocity.

[0098] In this embodiment of the invention, it is necessary to further explain that, based on the anti-interference benchmark feature library, the physical laws of electrolyte evaporation and plate corrosion are embedded into the topology update and feature fusion process of the graph neural network, so as to solve the misjudgment problem caused by the reliance on statistical correlation in traditional multimodal fusion.

[0099] The execution process of the topology update rule in a graph neural network includes:

[0100] The electrolyte evaporation process is modeled as a concentration diffusion equation, and the electrode corrosion process is modeled as an electrochemical reaction equation.

[0101] The physical equations for electrolyte evaporation and electrode corrosion include concentration diffusion equations and electrochemical reaction equations;

[0102] The concentration diffusion equation and the electrochemical reaction equation are combined and discretized into the ratio of the concentration gradient difference between nodes to the compensation rate. This ratio is then multiplied by the correlation function to generate the edge weight calculation rule, where the edge weight w is... ij It is obtained by multiplying the absolute value of the electrolyte concentration gradient difference to the compensated ultrasonic group velocity by the correlation function Φ(f);

[0103] Adjust the edge weights according to the actual process parameters to ensure that the message transmission path of the graph neural network conforms to the physical laws of material corrosion diffusion;

[0104] The implementation method of the differentiable frequency band selection network is as follows:

[0105] Input signal: Frequency domain characteristics of the ultrasonic time-domain signal and electromagnetic wave complex impedance spectrum acquired in step one;

[0106] Defect-sensitive frequency band extraction: Calculate the product of the absolute value of the frequency domain gradient of the ultrasonic amplitude spectrum and the absolute value of the frequency domain gradient of the electromagnetic impedance spectrum to generate the energy distribution ΔE(f) of the defect-sensitive frequency band;

[0107] Dynamic attention weight generation: The energy distribution of defect-sensitive frequency bands is normalized, and a frequency band attention mask is generated by filtering through a preset threshold to suppress noise interference in non-defect frequency bands;

[0108] The multi-scale feature fusion and node update method is as follows:

[0109] Input features: The baseline feature vector constructed in step one is used as the initial node features of the graph neural network;

[0110] Feature propagation rule: The feature update of node i depends on the weighted sum of features of its neighboring nodes j, with the weights being the edge weights w. ij The product of the frequency band value corresponding to the frequency band attention mask;

[0111] Output data: The updated node features are passed to step three for defect morphology inversion.

[0112] In one possible embodiment, the topology update rule includes an edge weight update formula, which is obtained as follows:

[0113] Let nodes i and j represent the indices of the feature vectors of different spatial locations inside the electrolytic capacitor; the edge weights are obtained using the following formula;

[0114]

[0115] Where, ΔC ij This represents the difference in electrolyte concentration gradient between nodes i and j;

[0116] The edge weights are summed to obtain the edge weight matrix.

[0117] In this embodiment of the invention, it needs to be further explained that the method for obtaining the multi-scale fusion features is as follows:

[0118] Wavelet transform is performed on ultrasonic time-domain signals to extract multi-resolution features (macroscale is used to capture overall corrosion, and microscale is used to locate micro-defects), and short-time Fourier transform is performed on electromagnetic wave complex impedance spectrum to obtain local frequency domain energy distribution.

[0119] Based on the frequency band attention mask, the ultrasonic wavelet coefficients and electromagnetic wave frequency domain energy are weighted and fused to enhance the signal strength in the defect-sensitive frequency band.

[0120] By using the message passing mechanism of graph neural networks, the weighted cross-modal features are aggregated in layers according to different scales (macro, meso, and micro). Combined with the topological constraints of the electrolyte diffusion path, spatial-frequency-temporal three-dimensional fusion features are generated, providing highly discriminative input for defect inversion.

[0121] In this embodiment of the invention, it needs to be further explained that the process of obtaining the three-dimensional defect morphology includes:

[0122] Input multi-scale fusion features, calculate the defect gradient field of the defect boundary through the spatial gradient of the multi-scale fusion features, and quantify the confidence of the defect boundary;

[0123] The encoder-decoder architecture is used to obtain the voxel mesh corresponding to the 3D defect morphology. The encoder compresses the fused features into a low-dimensional latent vector through a 3D convolutional layer to extract the macroscopic defect distribution. The decoder gradually upsamples through a 3D deconvolutional layer and updates the voxel mesh by combining the physical constraint graph topology. The adjacency matrix of the physical constraint graph topology serves as the spatial connection template for the decoder, which restricts the diffusion path of the defect morphology during the upsampling process to be consistent with the actual physical law of electrolyte corrosion. This forces a smooth transition of defect probabilities between adjacent spatial locations to avoid isolated noise points.

[0124] The ultrasonic wave propagation time difference and electromagnetic wave phase difference are calculated, and the calculation results drive the correction of the three-dimensional defect morphology. If the ultrasonic wave propagation time difference exceeds the threshold, it indicates that the material density in the defect area is reduced or there are voids. The voxel value (defect probability) at the corresponding position is increased. If the electromagnetic wave phase difference is lower than the requirement, it indicates that the conductivity of the electrolyte is reduced or the electrode is corroded. The boundary confidence of the corrosion area is enhanced.

[0125] In one possible embodiment, the weighted error between the ultrasonic wave propagation time difference and the electromagnetic wave phase difference is denoted as the counterfactual reasoning loss function, which is transmitted to the encoder through the deconvolution layer of the decoder and fed back to the inversion network to update the weights of the three-dimensional convolution kernel. In another possible embodiment, the weights of the first three layers of the encoder's convolution kernel are fixed, and only the parameters of the last two layers and the decoder are finely adjusted to prevent overfitting. The decoder parameters are adjusted so that the generated three-dimensional defect morphology gradually approximates the actual physical laws.

[0126] In this embodiment of the invention, it is necessary to further explain that the regularization coefficient is dynamically adjusted based on the confidence level of the defect gradient field, the non-equilibrium entropy yield, or a fusion of the two. The regularization coefficient is used to control the smoothing intensity.

[0127] In this embodiment of the invention, it is necessary to further explain that the regularization coefficient is dynamically adjusted based on the confidence level of the defect gradient field:

[0128] For high-confidence regions (defect gradient field confidence ≥ confidence threshold), reduce the regularization coefficient and prioritize preserving the sharp features of the defect boundary;

[0129] For low-confidence regions (defect gradient field confidence < confidence threshold), enhance the regularization coefficient to suppress artifact interference caused by noise;

[0130] The confidence threshold is adaptively set based on the gradient field statistical distribution of historical defect samples; after each batch of testing is completed, the threshold is dynamically updated based on the defect gradient field confidence distribution of the current sample.

[0131] In one possible embodiment, the mean μ and standard deviation σ of the defect gradient field confidence are calculated based on a historical defect sample library, with an initial threshold set to μ-2σ. The defect gradient field confidence is mapped to a probability value by the sigmoid function, which directly reflects the reliability of the defect boundary.

[0132] In this embodiment of the invention, it is necessary to further explain that the regularization coefficient is dynamically adjusted based on the non-equilibrium entropy yield.

[0133] Based on the electrolyte temperature field T(x, y, z) and the dissipation function σ(x, y, z), the formula is used to... Calculate the non-equilibrium entropy yield of the defect region. Non-equilibrium entropy yield represents the amount of energy dissipated in the defect region per unit time;

[0134] The electrolyte temperature field is measured by an infrared thermal imager or temperature sensor; the dissipation function is used to characterize the energy dissipation density of the material corrosion process; Ω represents the spatial extent of the defect region.

[0135] If the non-equilibrium entropy yield threshold is exceeded Increase the regularization coefficient to enhance smoothing strength and suppress noise interference in high dissipation regions;

[0136] If the non-equilibrium entropy yield threshold is not exceeded, the regularization coefficient is reduced to preserve defect details in the low-dissipation region.

[0137] In this embodiment of the invention, it is necessary to further explain that the fusion dynamic adjustment regularization coefficient based on the defect gradient field confidence and the non-equilibrium entropy yield includes:

[0138] By combining the confidence level of the signal defect gradient field with the non-equilibrium entropy yield, the limitations of a single dimension can be overcome.

[0139] Let the regularization coefficient be denoted as λ, λ0 represent the initial regularization coefficient, and T be the confidence level of the defect gradient field. z ,

[0140]

[0141] Among them, tanh(·) is used to compress the entropy production ratio to the (-1, 1) interval to avoid extreme values ​​causing the regularization coefficient to run out of control.

[0142] Summary: Example 1 achieved high-precision defect detection and process optimization of electrolytic capacitors by combining multimodal signal acquisition and deep learning technology. The specific working principle is as follows:

[0143] The multimodal physical coupling signal synchronous acquisition mechanism ensures the time and frequency synchronization of ultrasonic and electromagnetic wave signals, providing rich defect feature information, which solves the problem of insufficient feature extraction in single-modal detection methods. The physical constraint cross-modal modeling utilizes the physical properties of electrolyte and electrode plates to dynamically update the graph neural network structure, and extracts defect-sensitive features through a differentiable frequency band selection network, realizing the effective fusion of multi-source heterogeneous data and improving the accuracy of defect classification and localization. The edge-adaptive defect morphology inversion technology achieves high-precision three-dimensional defect morphology reconstruction through multi-scale feature fusion and edge-preserving regularization, further improving the accuracy of defect detection.

[0144] Background: The separation of the inspection process from the production process makes it impossible to dynamically optimize manufacturing parameters through real-time data feedback, resulting in a difficulty in systematically reducing the defect rate. This is because existing technologies cannot effectively feed inspection results back into the production process, failing to achieve closed-loop optimization of inspection and production. Based on this, the present invention provides the solution in Embodiment 2.

[0145] Example 2, the difference between this embodiment and Example 1, is that it further includes a production process optimization step, including:

[0146] Step 4: Causal-driven process closed-loop optimization: Construct a multi-objective decision model with defect confidence and material property drift as the observation states, and generate a Pareto optimal process solution set by modeling the causal relationship between process parameters and defect formation.

[0147] Furthermore, an adversarial training strategy is employed to jointly optimize the content of Implementation Examples 1 and 2. Through adversarial training, Implementation Example 1 is forced to learn more robust and generalized features, while Implementation Example 2 is trained to cope with more diverse production situations. This helps the system better handle unseen defect types or production conditions, improving the effectiveness and reliability of the entire quality control and process optimization process. In practical applications, an adversarial training framework is set up, where Implementation Example 1 (understood as the detection model) acts as the generator for adversarial training, attempting to generate defect patterns that are difficult for the process controller to correct. Implementation Example 2 (the process controller) acts as the discriminant in adversarial training, matching the optimal process parameter adjustment scheme for each detected defect. For example, Implementation Example 1 identifies a complex electrolyte distribution non-uniformity pattern that is difficult to correct through simple process parameter adjustments; Implementation Example 2 learns to solve the existing electrolyte distribution non-uniformity pattern by combining adjustments of multiple parameters (such as temperature, pressure, electrolyte injection rate, etc.). Through multiple rounds of training, the adversarial training framework can maintain high efficiency and stability when facing various complex production situations.

[0148] The explanation explains that through adversarial training, Example 1 is forced to learn more robust and generalized features, while the process controller is trained to cope with more diverse production situations. This approach helps the system better handle unseen defect types or production conditions, improving the effectiveness and reliability of the entire quality control and process optimization process. In practical applications, an adversarial training framework is designed to attempt to generate defect patterns that are difficult for the process controller to correct. The process controller, acting as a "discriminator," attempts to find the optimal process parameter adjustment scheme for each detected defect. For example, Example 1 might identify a complex electrolyte distribution non-uniformity pattern that is difficult to correct with simple process parameter adjustments. The process controller then needs to learn how to address this situation by combining fine adjustments of multiple parameters (such as temperature, pressure, electrolyte injection rate, etc.). Through multiple rounds of training, Example 1 becomes more sensitive to potential problem patterns, while the process controller is able to develop more refined and effective adjustment strategies, enabling the entire system to remain efficient and stable in the face of various complex production situations.

[0149] Working Principle: Example 2 establishes a correlation between defect detection results and production process parameters based on causal-driven closed-loop optimization. Optimal process parameters are generated through a multi-objective decision model, achieving closed-loop optimization of the detection process and production process, which helps to systematically reduce the defect rate. Through synergy with Example 1, it effectively solves the problems existing in the prior art, achieving efficient, high-precision, and cost-controllable defect detection and process optimization for electrolytic capacitors.

[0150] Example 3, see Figure 2 The invention provides a structural block diagram of an electrolytic capacitor testing system, comprising:

[0151] The test signal synchronous acquisition module synchronously acquires ultrasonic and electromagnetic wave signals through a phase-locked trigger acquisition mechanism, generates a time-frequency domain coupled waveform data stream, and transmits it to the dynamic compensation module;

[0152] The dynamic compensation module constructs an anti-interference benchmark feature library covering the process fluctuation range. Based on the ratio of the current physical property parameters to the benchmark electrolyte, it dynamically compensates for the ultrasonic propagation speed, calculates the corrected ultrasonic speed, and transmits it to the graph engine module; it also transmits the anti-interference benchmark feature library to the frequency selection network module.

[0153] The graph engine module dynamically generates the topology update rules of the graph neural network based on the physical equations of electrolyte evaporation and electrode corrosion, and outputs the graph topology structure with physical constraints.

[0154] The frequency selection network module is used to calculate the energy distribution of the defect-sensitive frequency band by multiplying the gradient across the modal frequency domain. After normalization, it generates a frequency band attention mask and transmits it to the fusion engine module.

[0155] The fusion engine module extracts multi-scale fusion features from time-frequency coupled waveforms based on defect-sensitive frequency band features, inputs the multi-scale fusion features into the edge-preserving regularization layer, adaptively adjusts the reconstruction parameters according to the confidence of the defect gradient field, generates a three-dimensional defect morphology that meets the continuity constraint, and transmits it to the inversion optimization module.

[0156] The inversion optimization module generates a voxel mesh through an encoder-decoder architecture and applies an edge-preserving regularization layer. It obtains a counterfactual inference loss function based on the weighted summation of the ultrasonic wave propagation time difference and the electromagnetic wave phase difference, and transmits it to the graph engine module to update the weights of the 3D convolution kernel, thereby achieving closed-loop feedback.

[0157] In conclusion, 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 of testing an electrolytic capacitor, characterized by, Includes the following steps: Step 1: Synchronous Acquisition of Multimodal Physically Coupled Signals: Establish a phase-locked trigger acquisition mechanism for ultrasonic and electromagnetic waves to generate time-frequency coupled waveforms of dual-mode signals; based on the correlation function between the electrolyte dynamic impedance spectrum and ultrasonic velocity... An anti-interference benchmark feature library covering the process fluctuation range was constructed, and the ultrasonic propagation velocity was dynamically compensated based on the ratio of the physical property parameters of the current electrolyte to the benchmark electrolyte. By traversing different gradient electrolyte formulations, encapsulation stress, and temperature combinations, benchmark feature vectors were collected. Each set of benchmark feature vectors includes a correlation function between the electrolyte dynamic impedance spectrum and the ultrasonic velocity. Extracting the frequency domain amplitude spectrum and electromagnetic impedance spectrum frequency domain gradient from the ultrasonic amplitude sequence using Fourier transform; based on the electrolyte dynamic impedance spectrum. With ultrasonic speed The relationship between them Let f represent the angular frequency, and f represent the frequency. The correlation function is calculated using the following formula. : The current electrolyte viscosity and dielectric constant are denoted as follows: , The viscosity and dielectric constant of the reference electrolyte are denoted as follows: , ; The dynamic compensation formula automatically corrects for the impact of process fluctuations on ultrasonic wave propagation speed, ensuring consistent testing conditions. The dynamic compensation formula for ultrasonic wave speed is as follows: in, This represents the corrected ultrasonic velocity; the corrected ultrasonic velocity is calculated using a dynamic compensation formula. Then, the ultrasonic speed Add these features to the corresponding benchmark feature vectors and compile them to obtain the anti-interference benchmark feature library; Step 2: Cross-modal modeling of physical constraints: Based on the physical equations of electrolyte evaporation and plate corrosion, the topology update rules of the graph neural network are dynamically generated, and the graph topology of physical constraints is output; the defect-sensitive frequency band features of the cross-modal signal are extracted through a differentiable frequency band selection network, and dynamic attention weights are generated. Step 3: Edge-adaptive defect morphology inversion: Based on the defect-sensitive frequency band characteristics, multi-scale fusion features are extracted from the time-frequency domain coupled waveform. The multi-scale fusion features are input into the edge-preserving regularization layer. The reconstruction parameters are adaptively adjusted according to the confidence of the defect gradient field to generate a three-dimensional defect morphology. Based on the difference in propagation characteristics between ultrasonic waves and electromagnetic waves, the physical consistency of the inversion results is optimized.

2. The method for testing electrolytic capacitors according to claim 1, characterized in that, Based on an anti-interference benchmark feature library, the physical laws governing electrolyte evaporation and electrode corrosion are embedded into the topology update and feature fusion process of a graph neural network. The operation process of the graph neural network's topology update rules includes: The electrolyte evaporation process is modeled as a concentration diffusion equation, and the electrode corrosion process is modeled as an electrochemical reaction equation. The physical equations for electrolyte evaporation and electrode corrosion include concentration diffusion equations and electrochemical reaction equations; The concentration diffusion equation and the electrochemical reaction equation are combined and discretized. The calculation rule for the edge weights is generated by multiplying the absolute value of the electrolyte concentration gradient difference between nodes i and j by the ratio of the corrected ultrasonic velocity to the correlation function; the edge weight w... ij The ratio of the absolute value of the electrolyte concentration gradient difference to the corrected ultrasonic velocity, multiplied by the correlation function... get; Adjust the edge weights according to the actual process parameters to ensure that the message transmission path of the graph neural network conforms to the physical laws of material corrosion diffusion; The implementation method of the differentiable frequency band selection network is as follows: Input signal: Frequency domain characteristics of ultrasonic time-domain signal and electromagnetic wave complex impedance spectrum; Defect-sensitive frequency band extraction: Calculate the product of the absolute value of the frequency domain gradient of the ultrasonic amplitude spectrum and the absolute value of the frequency domain gradient of the electromagnetic impedance spectrum to generate the energy distribution ΔE(f) of the defect-sensitive frequency band. Dynamic attention weight generation: The energy distribution of defect-sensitive frequency bands is normalized, and a frequency band attention mask is generated by filtering through a preset threshold to suppress noise interference in non-defect frequency bands; The multi-scale feature fusion and node update method is as follows: Input features: The constructed baseline feature vector serves as the initial node features of the graph neural network; Feature propagation rule: feature update of node i depends on the weighted sum of features of neighboring nodes j, with weights being edge weights w ij a product of the corresponding band attention mask and the band value; Output data: The updated node features are passed to step three for defect morphology inversion.

3. The electrolytic capacitor testing method according to claim 1, characterized in that, The method for obtaining multi-scale fusion features is as follows: Wavelet transform is performed on ultrasonic time-domain signals to extract multi-resolution features, and short-time Fourier transform is performed on electromagnetic wave complex impedance spectrum to obtain local frequency domain energy distribution; Based on the frequency band attention mask, the ultrasonic wavelet coefficients and electromagnetic wave frequency domain energy are weighted and fused to enhance the signal strength in the defect-sensitive frequency band. By using the message passing mechanism of graph neural networks, the weighted cross-modal features are aggregated in layers according to different scales, and combined with the topological constraints of electrolyte diffusion paths, spatial-frequency-temporal three-dimensional fusion features are generated, providing highly discriminative input for defect inversion.

4. The method for testing electrolytic capacitors according to claim 1, characterized in that, The process of obtaining the three-dimensional defect morphology includes: Input multi-scale fusion features, calculate the defect gradient field of the defect boundary through the spatial gradient of the multi-scale fusion features, and quantify the confidence of the defect boundary; The encoder-decoder architecture is used to obtain the voxel mesh corresponding to the 3D defect morphology. The encoder compresses the fused features into a low-dimensional latent vector through a 3D convolutional layer to extract the macroscopic defect distribution. The decoder upsamples step by step through a 3D deconvolutional layer and updates the voxel mesh in combination with the physical constraint graph topology. The adjacency matrix of the physical constraint graph topology serves as the spatial connection template of the decoder to limit the diffusion path of the defect morphology during the upsampling process, making it consistent with the actual physical law of electrolyte corrosion, and forcing a smooth transition of the defect probability in adjacent spatial positions to avoid isolated noise. The time difference of ultrasonic wave propagation and the phase difference of electromagnetic wave are calculated, and the calculation results are used to drive the correction of the three-dimensional defect morphology.

5. The electrolytic capacitor testing method according to claim 4, characterized in that, The regularization coefficient is dynamically adjusted based on the confidence level of the defect gradient field, the non-equilibrium entropy yield, or a fusion of the two. The regularization coefficient is used to control the smoothing intensity.

6. The method for testing electrolytic capacitors according to claim 5, characterized in that, Dynamically adjust the regularization coefficient based on the confidence level of the defect gradient field: For high-confidence regions, reduce the regularization coefficient and prioritize preserving the sharp features of defect boundaries; For low-confidence regions, enhance the regularization coefficient to suppress artifact interference caused by noise; The confidence threshold for the defect gradient field is adaptively set based on the statistical distribution of the gradient field of historical defect samples; after each batch of testing is completed, the threshold is dynamically updated based on the confidence distribution of the defect gradient field of the current sample.

7. The method for testing electrolytic capacitors according to claim 6, characterized in that, Dynamic adjustment of regularization coefficient based on non-equilibrium entropy yield: Based on the electrolyte temperature field With dissipation function Through formula Calculate the non-equilibrium entropy yield of the defect region. The non-equilibrium entropy yield represents the energy dissipation of the defect region per unit time; where x, y, and z represent the coordinates of the spatial sampling points within the spatial range Ω of the defect region in the three-dimensional coordinate system. The electrolyte temperature field is measured by an infrared thermal imager or temperature sensor; the dissipation function is used to characterize the energy dissipation density of the material corrosion process; Ω represents the spatial extent of the defect region. If the non-equilibrium entropy yield threshold is exceeded Increase the regularization coefficient to enhance smoothing strength and suppress noise interference in high dissipation regions; If the non-equilibrium entropy yield threshold is not exceeded, the regularization coefficient is reduced to preserve defect details in the low-dissipation region.

8. The method for testing electrolytic capacitors according to claim 7, characterized in that, The regularization coefficient is dynamically adjusted based on the fusion of defect gradient field confidence and non-equilibrium entropy yield, including: Let the regularization coefficient be denoted as , Let represent the initial regularization coefficient, and let denote the confidence level of the defect gradient field as . , in, Used to compress the entropy production ratio to the (-1, 1) range.

9. An electrolytic capacitor testing system for implementing the electrolytic capacitor testing method of claim 1, characterized in that, include: The test signal synchronous acquisition module synchronously acquires ultrasonic and electromagnetic wave signals through a phase-locked trigger acquisition mechanism, generates a time-frequency domain coupled waveform data stream, and transmits it to the dynamic compensation module; The dynamic compensation module constructs an anti-interference benchmark feature library covering the process fluctuation range. It dynamically compensates for the ultrasonic propagation speed based on the ratio of the physical property parameters of the current electrolyte to the benchmark electrolyte, calculates the corrected ultrasonic speed, and transmits it to the graph engine module. Transmit the anti-interference benchmark feature library to the frequency-selective network module; The graph engine module dynamically generates the topology update rules of the graph neural network based on the physical equations of electrolyte evaporation and electrode corrosion, and outputs the graph topology structure with physical constraints. The frequency selection network module is used to calculate the product of the absolute value of the frequency domain gradient of the ultrasonic amplitude spectrum and the absolute value of the frequency domain gradient of the electromagnetic impedance spectrum, generate the energy distribution ΔE(f) of the defect-sensitive frequency band, generate the frequency band attention mask after normalization, and transmit it to the fusion engine module. The fusion engine module extracts multi-scale fusion features from time-frequency coupled waveforms based on defect-sensitive frequency band features, inputs the multi-scale fusion features into the edge-preserving regularization layer, adaptively adjusts the reconstruction parameters according to the confidence of the defect gradient field, generates a three-dimensional defect morphology that meets the continuity constraint, and transmits it to the inversion optimization module. The inversion optimization module generates a voxel mesh through an encoder-decoder architecture and applies an edge-preserving regularization layer. It obtains a counterfactual inference loss function based on the weighted summation of the ultrasonic wave propagation time difference and the electromagnetic wave phase difference, and transmits it to the graph engine module to update the weights of the 3D convolution kernel, thereby achieving closed-loop feedback.

Images