Aluminum electrolytic capacitor state evaluation method and device based on multi-dimensional data analysis

Abstract

The invention relates to the field of capacitor state evaluation, in particular to an aluminum electrolytic capacitor state evaluation method and device based on multi-dimensional data analysis. The method comprises the following steps: performing multi-band excitation on an aluminum electrolytic capacitor, and performing three-dimensional fault distribution reconstruction and structure degradation trend prediction so as to construct an impedance tomography degradation trend prediction map; collecting audio signals in the charging and discharging process of the capacitor, carrying out acoustic texture analysis, carrying out voiceprint-structure degradation correlation mining according to the impedance tomography degradation trend prediction map, and constructing an acoustic texture damage evolution evaluation model; obtaining a thermal imaging image of the capacitor, carrying out local overheating diffusion mining, and constructing a three-dimensional thermal anomaly detection model; performing electrolyte chemical component change tracking and electrolyte degradation situation analysis on the electrolyte of the capacitor, and constructing an electrolyte degradation situation curve; according to the invention, efficient and accurate capacitor state evaluation is realized. The operation reliability and stability of capacitor equipment are improved, and the service life of the equipment is effectively prolonged.

Description

Technical Field

[0001] The present invention relates to the field of capacitor status assessment, and in particular to a method and device for assessing the status of an aluminum electrolytic capacitor based on multi-dimensional data analysis. Background Art

[0002] As electronic devices increasingly advance toward high performance and high integration, aluminum electrolytic capacitors (AECs), key energy storage and filtering components in power systems, are widely used in various industrial control systems, power electronics, communications equipment, and new energy vehicles. During long-term operation in harsh environments such as high temperature, high frequency, and high voltage, aluminum electrolytic capacitors are susceptible to physical and chemical factors such as internal electrolyte volatilization, oxide film aging, and electrode corrosion, leading to gradual performance degradation. In severe cases, this can cause system failures or safety incidents. Therefore, accurate and real-time assessment of the health of aluminum electrolytic capacitors is crucial for ensuring system stability and safety.

[0003] Traditional methods for assessing the condition of aluminum electrolytic capacitors rely primarily on regular maintenance and manual inspections, typically based on simple electrical parameter tests (such as ESR, capacitance change, and leakage current) to determine performance degradation. These methods not only rely on the experience of professionals, have long inspection cycles, and are inefficient, but they also often fail to fully monitor the equipment during actual operation, easily overlooking potential faults. Furthermore, traditional methods lack comprehensiveness and predictive power when addressing the many complex factors that influence capacitor performance, making them unable to meet the real-time, intelligent, and refined management requirements of modern, complex applications. Summary of the Invention

[0004] In order to solve the above technical problems, the present invention proposes a method and device for evaluating the status of aluminum electrolytic capacitors based on multi-dimensional data analysis to solve at least one of the above technical problems.

[0005] To achieve the above objectives, the present invention provides a method for evaluating the status of an aluminum electrolytic capacitor based on multi-dimensional data analysis, comprising the following steps: Step S1: performing multi-band excitation on the aluminum electrolytic capacitor, and performing three-dimensional fault distribution reconstruction and structural degradation trend prediction, thereby constructing an impedance tomography degradation trend prediction map; Step S2: Collect audio signals during the capacitor's charging and discharging process, perform acoustic texture analysis, and mine the soundprint-structure degradation correlation based on the impedance tomography degradation trend prediction diagram to construct an acoustic texture damage evolution assessment model. Step S3: Acquire thermal imaging images of the capacitor, perform spatiotemporal evolution analysis of temperature distribution, conduct local overheating diffusion mining, and construct a three-dimensional thermal anomaly detection model; Step S4: Tracking the chemical composition changes of the capacitor electrolyte and analyzing the electrolyte degradation trend, and constructing an electrolyte degradation trend curve; Step S5: Based on the acoustic texture damage evolution assessment model, the three-dimensional thermal anomaly detection model, and the electrolyte degradation situation curve, multi-dimensional feature space mapping and fuzzy logic reasoning between features are performed to obtain fuzzy assessment results between multi-dimensional features; Step S6: performing capacitor failure risk level assessment based on the fuzzy assessment result and capacitor performance degradation prediction to perform capacitor status assessment.

[0006] In this specification, a device for evaluating the state of an aluminum electrolytic capacitor based on multi-dimensional data analysis is provided, which is used to perform the method for evaluating the state of an aluminum electrolytic capacitor based on multi-dimensional data analysis as described above, comprising: The impedance tomography module is used to perform multi-band excitation on aluminum electrolytic capacitors, reconstruct three-dimensional fault distribution, and predict structural degradation trends, thereby constructing an impedance tomography degradation trend prediction map; The acoustic texture module is used to collect audio signals from the capacitor's charging and discharging processes, perform acoustic texture analysis, and mine the soundprint-structure degradation correlation based on the impedance tomography degradation trend prediction chart to build an acoustic texture damage evolution assessment model. The temperature distribution module is used to obtain thermal imaging images of capacitors, analyze the spatiotemporal evolution of temperature distribution, and conduct local overheating diffusion mining to build a three-dimensional thermal anomaly detection model; Degradation trend analysis module, used to track changes in electrolyte chemical composition and analyze electrolyte degradation trends of capacitor electrolytes, and construct electrolyte degradation trend curves; The feature space mapping module is used to perform multi-dimensional feature space mapping and fuzzy logic reasoning between features based on the acoustic texture damage evolution assessment model, the three-dimensional thermal anomaly detection model, and the electrolyte degradation situation curve to obtain fuzzy evaluation results between multi-dimensional features; The comprehensive status assessment module is used to assess the capacitor failure risk level according to the fuzzy assessment result and predict the capacitor performance attenuation to perform the capacitor status assessment operation.

[0007] The beneficial effects of the present invention are as follows: by applying multi-band electrical excitation to the aluminum electrolytic capacitor and reconstructing its three-dimensional impedance fault distribution, it is possible to extract the internal structural degradation characteristics from the frequency domain response; combining the spatial evolution law of the fault distribution to construct an impedance tomography degradation trend prediction map, compared with the traditional single-point impedance value determination method, it can accurately quantify the degradation intensity and expansion path of different structural areas, effectively supporting early fault detection and regional health identification. Introducing an acoustic signal acquisition mechanism during the capacitor charging and discharging process to capture the acoustic texture characteristics formed by weak phenomena such as dielectric breakdown, bubbling, and partial discharge in the audio; combining the impedance tomography degradation map, deeply coupling the voiceprint data with the internal structural changes for analysis, and constructing an acoustic texture damage evolution model to achieve contactless and non-invasive dynamic health perception of the capacitor, thereby improving the detection dimension and accuracy. Thermal imaging is used to model the spatiotemporal evolution of capacitor temperature distribution, identifying "hotspot drift" and "thermal field asymmetry" during operation. Localized overheating diffusion analysis is then performed to construct a three-dimensional thermal anomaly detection model. This model significantly improves the ability to identify abnormal heat diffusion paths under high-load conditions, preventing sudden damage caused by accumulated localized temperature rise. By tracking the chemical composition changes within the capacitor's electrolyte in real time and combining them with environmental factors such as operating time and current, an electrolyte degradation curve is derived, enabling digital modeling of the aging mechanism at the material level. This curve reflects the entire process of capacitor failure, from degradation to sub-health to critical failure, contributing to the development of a full-cycle prediction capability for material degradation. The multimodal data (acoustic texture, thermal images, electrolyte characteristics, etc.) is mapped into a unified multidimensional feature space, and a fuzzy logic inference mechanism is introduced to handle the uncertain coupling relationships between the features. This mechanism avoids misjudgments caused by fluctuations in a single indicator, and produces comprehensive, fault-tolerant, and structurally aware fuzzy health assessment results. Based on the fuzzy assessment results, capacitor failure risk levels are classified and performance degradation trends are predicted, achieving fine-grained capacitor status classification, risk visualization, and full life cycle status evolution assessment. This module supports intelligent manufacturing and predictive maintenance tasks, realizing the transformation from "static detection" to "dynamic prediction," significantly improving the operational reliability and remaining life estimation capabilities of capacitors in critical electronic systems. The entire solution integrates electrical, acoustic, thermal, and chemical multi-dimensional sensing channels to build a full-dimensional status assessment system from structural distribution and operating behavior to material degradation, breaking through the technical bottleneck of traditional capacitor detection that relies solely on electrical parameters and thermal protection. It achieves high-throughput, non-contact, process-based, and multi-source integrated status assessment capabilities, providing comprehensive technical support for the health management of core capacitors in high-reliability power systems, industrial equipment, and power electronics scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] Figure 1 This is a schematic flow chart of the steps of a method for evaluating the state of an aluminum electrolytic capacitor based on multi-dimensional data analysis according to the present invention; Figure 2 Detailed implementation flow chart of step S1; Figure 3 Detailed implementation flow chart of step S2; Figure 4 Schematic diagram of the detailed implementation steps of step S3. DETAILED DESCRIPTION

[0009] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0010] This application provides a method and device for assessing the condition of an aluminum electrolytic capacitor based on multi-dimensional data analysis. The execution entities of the method and device for assessing the condition of an aluminum electrolytic capacitor based on multi-dimensional data analysis include, but are not limited to, the following: mechanical equipment, data processing platforms, cloud server nodes, network upload devices, etc., which can be considered as general computing nodes of this application. The data processing platform includes, but is not limited to, at least one of an audio and image management system, an information management system, and a cloud data management system.

[0011] See also Figures 1 to 4 The present invention provides a method for evaluating the state of an aluminum electrolytic capacitor based on multi-dimensional data analysis, comprising the following steps: Step S1: performing multi-band excitation on the aluminum electrolytic capacitor, and performing three-dimensional fault distribution reconstruction and structural degradation trend prediction, thereby constructing an impedance tomography degradation trend prediction map; Step S2: Collect audio signals during the capacitor's charging and discharging process, perform acoustic texture analysis, and mine the soundprint-structure degradation correlation based on the impedance tomography degradation trend prediction diagram to construct an acoustic texture damage evolution assessment model. Step S3: Acquire thermal imaging images of the capacitor, perform spatiotemporal evolution analysis of temperature distribution, conduct local overheating diffusion mining, and construct a three-dimensional thermal anomaly detection model; Step S4: Tracking the chemical composition changes of the capacitor electrolyte and analyzing the electrolyte degradation trend, and constructing an electrolyte degradation trend curve; Step S5: Based on the acoustic texture damage evolution assessment model, the three-dimensional thermal anomaly detection model, and the electrolyte degradation situation curve, multi-dimensional feature space mapping and fuzzy logic reasoning between features are performed to obtain fuzzy assessment results between multi-dimensional features; Step S6: performing capacitor failure risk level assessment based on the fuzzy assessment result and capacitor performance degradation prediction to perform capacitor status assessment.

[0012] In the embodiment of the present invention, see Figure 1, is a schematic flow chart of the steps of a method for evaluating the status of an aluminum electrolytic capacitor based on multi-dimensional data analysis of the present invention. In this example, the steps of the method for evaluating the status of an aluminum electrolytic capacitor based on multi-dimensional data analysis include: Step S1: performing multi-band excitation on the aluminum electrolytic capacitor, and performing three-dimensional fault distribution reconstruction and structural degradation trend prediction, thereby constructing an impedance tomography degradation trend prediction map; In this embodiment, a multi-channel impedance analyzer (such as the Solartron 1260) was used to excite the aluminum electrolytic capacitor under test during the multi-band excitation and data acquisition phase. The frequency range covered 10 Hz to 1 MHz, with logarithmic sampling intervals. A total of 50 frequency points were used to obtain full-band response data from low frequencies (reflecting polarization effects) to high frequencies (reflecting structural tightness). The excitation voltage at each frequency point was set to 100 mV RMS to ensure sufficient signal-to-noise ratio without causing damage. In the experiment, aluminum electrolytic capacitor samples with different usage cycles (for example, 1000 h, 3000 h, 5000 h, and new capacitors) were selected to obtain electrical response data at different degradation stages. During the excitation process, impedance modulus, phase angle, real part, and imaginary part data were collected from the capacitor pins. Data collection was performed using a high-precision synchronous sampling card (sampling rate exceeding 500 kHz) to ensure both time and frequency accuracy.

[0013] An impedance tomography reconstruction algorithm based on the finite element method (FEM) is used to fit the spatial distribution of multi-band responses. Due to the internal structure of aluminum electrolytic capacitors, traditional two-dimensional cross-sectional reconstruction is insufficient to fully reflect their degradation process. Therefore, a three-dimensional volumetric reconstruction model is used. A full-scale geometric model of the capacitor is constructed, and the acquired frequency response data is mapped to different locations on the model. A multi-layer Bayesian inference framework is used to estimate the spatial variations of the conductivity and dielectric constant. During the solution process, regularization optimization (such as Tikhonov regularization) is used to suppress ill-posed solutions caused by measurement noise. The output of this stage is a three-dimensional distribution map reflecting the changes in the capacitor's internal electrical properties at different frequencies. Each layer of the image represents the local structural features in a frequency band. Reconstructed images from different time samples are compared and analyzed. By extracting key features from the image (such as the mean conductivity of the electrode surface and the dielectric constant gradient at the aluminum foil-electrolyte interface), a feature vector sequence is constructed. The degradation evolution process is then predicted using time series modeling methods (such as long short-term memory networks (LSTMs) or Kalman filtering). Experiments have shown that the dielectric constant within capacitors shows a clear downward trend with increasing usage, and "hot spots" gradually appear in the conductivity distribution, indicating localized degradation of the dielectric or electrode materials. Based on these results, an impedance tomography degradation trend prediction map can be constructed, showing the multi-frequency, multi-timepoint fault evolution process from the initial state to the estimated future state, which can be used to quantitatively assess the lifespan and health of capacitors.

[0014] Step S2: Collect audio signals during the capacitor's charging and discharging process, perform acoustic texture analysis, and mine the soundprint-structure degradation correlation based on the impedance tomography degradation trend prediction diagram to construct an acoustic texture damage evolution assessment model. In this embodiment, a high-sensitivity electret condenser microphone (such as the Knowles SPU0410) is paired with a low-noise preamplifier. The sampling frequency is set to 96 kHz, and the quantization accuracy is 24 bits to ensure the ability to capture weak signals. The capacitor is tested in a relatively soundproof test chamber to eliminate external noise interference. The charge and discharge processes are driven by a controllable constant voltage source, with a typical charging voltage set at 25 V. The discharge process releases energy through a parallel load resistor (10 Ω). The test cycle consists of 50 charge and discharge cycles per capacitor sample to measure the stability and repeatability of the acoustic signal. The collected raw sound data undergoes preliminary preprocessing, including filtering (bandpass range 2-40 kHz), normalization, and denoising to ensure the accuracy of subsequent feature extraction. By framing and windowing the time series acoustic signal (frame length 50 ms, frame shift 25 ms), various typical acoustic features are extracted, including Mel-frequency cepstral coefficients (MFCCs), spectral centroid, short-term energy variation, zero-crossing rate, and spectral flatness. The MFCC feature dimension is set to 13, and its first- and second-order differences are combined to form a complete voiceprint expression matrix. Principal component analysis (PCA) or linear discriminant analysis (LDA) methods are used to reduce the dimensionality of high-dimensional acoustic texture features and extract highly discriminative acoustic feature factors. Furthermore, audio data is converted into an image representation by constructing time-frequency graphs (such as short-time Fourier transform (STFT) images or Mel spectrograms). A convolutional neural network (CNN) model is then introduced for deep pattern recognition to detect audio texture anomalies caused by structural damage.

[0015] For each capacitor sample, the three-dimensional impedance tomography plots from step S1 were paired with their corresponding acoustic texture features to identify the corresponding relationship between acoustic response and physical structural degradation. Multivariate regression analysis and cross-validation were used to construct a nonlinear mapping relationship between structural damage and acoustic print features. If the sample size permits, random forest or support vector regression (SVR) models can be further introduced to model the coupling between acoustic features and structural parameters (such as the mean square error of the conductivity distribution and the local gradient of the dielectric constant). In particular, spectral analysis reveals that the corresponding frequencies of the slight "pops" or "sharp discharge sounds" produced by capacitors during the initial charging phase are concentrated between 15 and 30 kHz, highly correlated with the non-uniformly degraded regions at the electrode boundaries in the tomography plots, further validating the structural sensitivity of the acoustic signal. During the construction of the acoustic texture damage evolution assessment model, the acoustic print evolution trajectories of capacitors at multiple time points and different aging states were integrated with the tomographic evolution trends to construct a degradation discrimination model based on temporal acoustic prints. This model supports predictive capabilities for unseen future samples. It can input audio data recorded during a capacitor's charging process and output the sample's likely degradation level or estimated remaining lifespan. Experiments have shown that when a mild disturbance occurs within the capacitor, its acoustic texture characteristics change in advance, demonstrating its foresight and potential as an early warning signal for structural degradation.

[0016] Step S3: Acquire thermal imaging images of the capacitor, perform spatiotemporal evolution analysis of temperature distribution, conduct local overheating diffusion mining, and construct a three-dimensional thermal anomaly detection model; In this example, a medium-wave infrared thermal imager (such as the FLIR A655sc) was used. Its thermal sensitivity is better than 0.05°C, its spatial resolution reaches 640×480 pixels, and its frame rate is set at 50 Hz. This allows for continuous tracking and recording of transient temperature changes during the capacitor's charge and discharge cycles. To ensure accurate acquisition, a temperature measurement platform was constructed in a constant temperature controlled environment (room temperature 25°C ± 1°C). A matte black coating with high emissivity (ε ≈ 0.95) was uniformly applied to the capacitor surface to eliminate temperature measurement errors caused by material reflectivity differences. The capacitor's operating voltage was set to 20 V, and it was frequently charged and discharged for 20 minutes. During this time, thermal image data was continuously collected, with one frame recorded every 5 seconds. A total of approximately 240 frames were recorded, forming a thermal image sequence dataset.

[0017] A thermal field analysis method based on image processing is used to preprocess the collected thermal image sequence, including background temperature correction, image registration (taking into account slight viewing angle drift), and noise removal. The statistical characteristics of the temperature distribution in each frame of the image (such as maximum temperature, average temperature, local temperature gradient, etc.) are then extracted, and a surface diagram of the two-dimensional temperature distribution evolving over time is constructed. By analyzing the trajectory of the temperature surface change, the starting point of the temperature rise, the heat propagation path, and the stable or abnormal temperature rise areas can be clearly identified, thereby reflecting the behavior of the heat source inside the capacitor. Typical phenomena include: when leakage occurs inside the capacitor or the contact resistance increases, an asymmetric temperature rise pattern will appear on the surface, and the maximum temperature point has a continuous migration characteristic. These appear as "hot spot" drift trajectories in the time series thermal map.

[0018] Based on spatial gradient calculation and edge detection algorithms (such as the Sobel or Laplacian operator), regions of sudden temperature changes within each thermal image frame are automatically identified, marking the boundaries of localized high-temperature areas. Subsequently, a heat diffusion model (such as a numerical solution to the two-dimensional Fourier heat conduction equation) is used to simulate the temperature field diffusion path, further evaluating the spatiotemporal evolution of the thermal anomaly from a point source to the surrounding area. To achieve three-dimensional representation, the thermal image sequence is stacked in the time dimension to construct a three-dimensional tensor data model of the temperature variation in space (x, y) and time (t). Based on this, volume rendering techniques or isothermal surface extraction algorithms (such as Marching Cubes) are introduced to reconstruct a three-dimensional dynamic model of the heat diffusion process, visualizing and refining the localized overheating behavior in three dimensions. If a region repeatedly exhibits a rapid and irreversible trend of temperature rise across multiple tests, it may be a "hotspot" caused by a structural defect, which can serve as an important indicator for determining whether to repair or replace the device. Historical thermal image data is correlated and compared with the obtained impedance tomography to determine the spatial overlap between the thermal anomaly area and the area of ​​electrical degradation. By fusing heat map features extracted by a convolutional neural network (CNN) with impedance map features, a multimodal classification model (such as a deep neural network (DNN) fused with feature vectors) is trained to automatically identify thermal anomaly patterns and assess degradation levels. The detection model is also deployed in an online monitoring system, enabling real-time identification of thermal anomaly trends during normal capacitor operation and providing risk level indicators (e.g., normal, mild anomaly, severe overheating). Experiments have demonstrated that this thermal detection model is highly sensitive to faults such as local breakdown and loose electrode contact, providing early warning information before capacitor performance significantly degrades, significantly improving the reliability and safety of system operations.

[0019] Step S4: Tracking the chemical composition changes of the capacitor electrolyte and analyzing the electrolyte degradation trend, and constructing an electrolyte degradation trend curve; In this example, aluminum electrolytic capacitor samples after different usage cycles (e.g., 0 h, 1000 h, 3000 h, and 6000 h) were selected for comparative analysis. To avoid volatilization or contamination of components during sample disassembly, a low-temperature seal-breaking technique was used to cut the outer shell in an inert atmosphere (nitrogen) and quickly extract the internal electrolyte liquid. The extracted electrolyte was transferred to a micro-sealed bottle via a quantitative syringe and refrigerated for future use. The extraction volume was controlled at 50–100 μL each time, and a quartz tube was used as a sample container to accommodate subsequent spectroscopic and chromatographic analysis instruments.

[0020] During the electrolyte chemical composition analysis phase, a variety of complementary chemical analysis techniques are used to qualitatively and quantitatively test electrolyte samples. The main analytical methods include gas chromatography-mass spectrometry (GC-MS), Fourier transform infrared spectroscopy (FTIR), ion chromatography (IC), and electrospray ionization mass spectrometry (ESI-MS). GC-MS is used to detect volatility changes in low molecular weight organic solvents, such as solvent decomposition products such as ethylene glycol and γ-butyrolactone. FTIR is used to analyze changes in functional groups in the molecular structure, such as the conversion of carboxyl, ester, and amine groups, revealing the oxidation or hydrolysis pathway of the electrolyte. IC technology is used to analyze the concentration of dissolved ions, such as nitrate, chloride, and ammonium, reflecting the changing trends of the conductive components in the electrolyte. ESI-MS can be used to identify changes in the aggregation state of complex organic anions or cations, such as the formation of multivalent ion clusters, which has a significant impact on later performance degradation.

[0021] Dynamic characteristic curves of electrolyte composition changes over time are constructed. Using the concentration change or absorption peak intensity of characteristic components as the vertical axis and the usage cycle as the horizontal axis, a set of typical degradation indicator curves are generated. For example, the γ-butyrolactone content gradually decreases with increasing usage, indicating significant thermal decomposition in high-temperature environments. Simultaneously, the intensity of medium-mass fragment peaks in the mass spectrum increases, indicating an increase in the concentration of oxidation byproducts. These indicators together form a profile of electrolyte degradation. Furthermore, principal component analysis (PCA) is used to extract key component degradation characteristic variables, and regression models (such as polynomial fitting or exponential decay models) are used to model their changing trends. This results in a quantitative "electrolyte degradation trend curve" that can be used to predict the extent of electrolyte quality decline over future operating cycles. During the correlation and validation analysis phase, the electrolyte composition change data is compared with the electrical performance indicators obtained in steps S1-S3 (such as impedance tomography, electroacoustic response, and thermal anomaly data) to assess the role of chemical degradation in driving physical performance degradation. Experimental data show that the increase in water content in the electrolyte is highly correlated with the rise in ESR, and there is a linear relationship between γ-butyrolactone degradation and the increase in capacitor leakage current. These results verify the direct coupling mechanism between changes in electrolyte composition and functional degradation.

[0022] Step S5: Based on the acoustic texture damage evolution assessment model, the three-dimensional thermal anomaly detection model, and the electrolyte degradation situation curve, multi-dimensional feature space mapping and fuzzy logic reasoning between features are performed to obtain fuzzy assessment results between multi-dimensional features; In this embodiment, the key features output by the three sub-models are standardized to facilitate unified mapping of the feature layer. For example, for the acoustic texture model, metrics such as the rate of change of the sound spectrum center of mass (in Hz / ms), the temporal stability index of the MFCC feature (in dimensionless units), and the acoustic energy density anomaly coefficient (in dB) are extracted. For the thermal anomaly model, metrics such as the maximum hot spot temperature rise gradient (in °C / s), the hot spot area expansion rate (in mm² / s), and the anomaly duration (in seconds) are extracted. For the electrolyte model, metrics such as the rate of change of key component concentrations (e.g., the rate of decrease of γ-butyrolactone (in μg / h)) and the change in the acidic / alkaline ion ratio (in dimensionless units) are collected. All metrics are converted to dimensionless values ​​between 0 and 1 using Z-score or Min-Max normalization to ensure comparability of different physical quantities. During the fuzzy membership function design phase, corresponding membership functions are constructed for the key features of each dimension based on expert knowledge and historical experimental data. Taking the acoustic energy density anomaly coefficient as an example, three fuzzy sets are defined: "low risk," "medium risk," and "high risk." Their membership functions can be expressed using trapezoidal or Gaussian functions. Based on experience, the duration of thermal anomalies is set as "normal" for less than 5 seconds, "suspicious" for 5–15 seconds, and "abnormal" for more than 15 seconds. The electrolyte concentration change rate is graded based on component stability. This fuzzy mapping approach facilitates the transition from physical indicators to cognitive "risk levels." A Mamdani-based fuzzy inference model is employed, with the input being the fuzzy grade output corresponding to the aforementioned multidimensional features. The inference rule base is constructed using empirical rules and experimental validation. For example, typical rules include: "If the acoustic characteristics are high risk, the thermal diffusion rate is high risk, and the electrolyte composition change is medium risk, then the overall degradation state is severe"; and "If the acoustic characteristics are medium risk, the thermal anomaly is normal, and the electrolyte is high risk, then the overall degradation state is moderate." A total of 60 fuzzy inference rules are set to ensure coverage of all typical combination scenarios. The inference output is a fuzzy set (e.g., "good condition," "mildly degraded," "moderately degraded," and "severely degraded"), which is then clarified using the centroid method to produce a final health index (HI) ranging from 0 to 1, with lower values ​​indicating worse condition. A cross-validation model was performed using a representative sample library of aluminum electrolytic capacitors, including samples in the initial state (0 h), mildly aged (1000 h), moderately aged (3000 h), and critically failed (over 6000 h). Voiceprint data, thermal image sequences, and electrolyte composition information were collected from these samples and fed into a fusion model for condition assessment. The assessment results were compared with actual performance test results (e.g., increased leakage current, increased ESR, and decreased capacitance) to evaluate the accuracy and robustness of the fuzzy assessment model.Experimental results show that the multidimensional fusion model has higher recognition accuracy in the fuzzy area of ​​state division, especially the ability to provide early warning for minor anomalies. Its average misjudgment rate is less than 8%, showing good practical value in capacitor preventive maintenance scenarios.

[0023] Step S6: performing capacitor failure risk level assessment based on the fuzzy assessment result and capacitor performance degradation prediction to perform capacitor status assessment.

[0024] In this embodiment, the fuzzy evaluation result of the received output is between 0 and 1, reflecting the comprehensive evaluation of the current state of the capacitor. Based on a large number of sample verifications and failure statistics, the HI value is divided into five risk level intervals, namely: healthy (HI ≥ 0.85), slightly degraded (0.70 ≤ HI<0.85), moderately degraded (0.50 ≤ HI<0.70), severely degraded (0.30 ≤ HI<0.50), and critical failure (HI<0.30). Each level of risk is equipped with corresponding processing suggestions. For example, the "healthy" level only requires regular monitoring, the "moderately degraded" level recommends strengthening monitoring and starting to prepare a replacement plan, and the "critical failure" level recommends immediate shutdown and replacement to prevent potential functional failures or safety accidents. In order to ensure the accuracy of risk classification, the system cross-validates the HI value with traditional performance parameters. For example, when the measured capacitance dropped by more than 20%, the leakage current increased by more than 300%, and the equivalent series resistance (ESR) increased by more than twice the initial value, the results were highly consistent with those for capacitor samples with an HI < 0.50, verifying the close correspondence between the fuzzy assessment results and actual performance degradation. The experiment selected 30 groups of capacitor samples, with usage cycles ranging from 0 to 7000 hours. Their HI values ​​and actual performance parameters were recorded at different aging stages. A nonlinear mapping relationship between HI and key performance degradation was established, enhancing the reliability of the model.

[0025] Based on the historical trajectory of risk levels and the evolution trends of multi-dimensional features, the future health trajectory of capacitors is predicted. The specific implementation method involves constructing a time-series-based health index evolution model, using modeling approaches such as ARIMA, LSTM neural networks, and exponential decay models. In an actual experiment, the LSTM model was trained to predict the health index (HI) values ​​of 20 capacitors. The input was the historical HI values ​​of each capacitor at five consecutive time points, along with the change rates of its acoustic, thermal, and electrolyte characteristics. The output was a forecast of the HI values ​​for the next one to three months. The results showed a prediction accuracy exceeding 95%, with a mean prediction error of less than ±0.04, demonstrating good trend retention. To enhance the model's interpretability, the prediction results also include the "contribution weight" of each key feature to the future HI value. For example, in a high-temperature aging scenario, changes in thermal anomaly indicators contribute over 55% to the decline in health index, while under high-frequency operating conditions, the degradation rate of acoustic texture becomes the dominant factor. This interpretability of feature weights helps operations and maintenance personnel accurately identify the dominant degradation mechanism and implement targeted management measures. During the design phase of the condition assessment result execution and feedback mechanism, all assessment results are integrated into the equipment management system through the condition assessment interface, generating an automated assessment report that includes information on the current health level, future performance trend forecasts, recommended treatment strategies, and analysis of major degradation factors. Furthermore, the system supports a threshold alarm mechanism. If the HI value drops sharply over a short period of time (e.g., ΔHI > 0.15 / month), a risk warning will be triggered, prompting relevant personnel to review the data and assess whether there are any sudden degradation factors (such as a sudden increase in ambient temperature or a sudden change in workload).

[0026] In this embodiment, refer to Figure 2 , is a flowchart of the detailed implementation steps of step S1. In this embodiment, the detailed implementation steps of step S1 include: Define a multi-band impedance measurement excitation source, apply multi-band excitation to aluminum electrolytic capacitors, perform omnidirectional impedance synchronous scanning based on a ring electrode array, and extract omnidirectional multi-band capacitor excitation response parameters; Calculating impedance values ​​of the capacitance excitation response parameters in different directions, performing spatial distribution fitting, and constructing an impedance distribution space coordinate mapping network; The three-dimensional fault distribution is reconstructed based on the impedance distribution space coordinate mapping network to construct a multi-spectral impedance three-dimensional tomographic image; The structural degradation trend is predicted based on the multi-spectral impedance three-dimensional tomography images, thereby constructing the impedance tomography degradation trend prediction map.

[0027] In this embodiment, a programmable excitation source (such as the Keysight E4980A) is used to output a multi-band sinusoidal signal with a frequency range of 100 Hz to 1 MHz, encompassing a total of 20 frequency points, fully covering the common electrical characteristic response range of aluminum electrolytic capacitors. To obtain a comprehensive data response, a circular electrode array is wrapped around the capacitor. This configuration consists of 16 pairs of electrodes arranged at equal intervals. Each pair of electrodes is excited and sampled by a multi-channel impedance analyzer. The system rotates between different electrode pairs at each frequency point as the excitation source, with the remaining electrode pairs acting as receivers, completing a complete set of omnidirectional scanning measurements. Each excitation cycle is set to 10 ms. The measurement is repeated five times at each frequency point and the average is taken to suppress noise interference. Hardware phase-locked detection is used to synchronously extract the real (resistance) and imaginary (reactance) components of the response signal. The resulting raw data matrix is: [frequency × electrode combination × response parameter (R, X)], which is the "multi-band capacitor excitation response parameter." After acquiring the raw multi-frequency response parameters, this step converts the data from the electrode array domain to the spatial domain, using numerical modeling to estimate the impedance spatially at different locations within the capacitor. This is achieved using finite element inverse modeling and spatial interpolation fitting. First, a three-dimensional finite element model is established with the electrode array as the boundary condition, treating the capacitor as the internal medium of an irregular ellipsoid. Using the response values ​​of each pair of excitation / receiving electrode combinations as boundary inputs, a multi-objective least squares inversion algorithm based on Tikhonov regularization is used to solve for the complex impedance of each discrete element within the capacitor (e.g., a 2 mm³ voxel). To further refine the spatial distribution characteristics, the inverted impedance data is subjected to three-dimensional interpolation fitting, using RBF (3D radial basis functions) or 3D spline interpolation methods to generate a continuous impedance distribution function across the entire spatial grid. This spatial function, Z(x, y, z, f), is the "impedance distribution spatial coordinate mapping network," which associates the impedance response value at a specific frequency at each spatial location, achieving a spatial mapping from the electrode response to the internal impedance state of the capacitor. At each frequency point, Z (x, y, z) data slices are extracted to form multi-frequency tomographic slices. Using the frequency axis as the "spectral dimension," images at different frequencies are superimposed to form a sequence of multispectral impedance slices. Subsequently, 3D reconstruction algorithms (such as MarchingCubes or Volume Rendering) are used to synthesize the slices into a 3D image volume, resulting in a complete multispectral impedance 3D tomographic image. Image enhancement utilizes pseudo-color mapping, mapping impedance amplitude differences at different frequencies to color channels (e.g., blue for low frequencies and red for high frequencies) to improve image clarity.In addition, Structure Tensor Analysis is used to enhance possible boundaries, faults or abnormal areas in the image, giving the image stronger structural recognition capabilities.

[0028] The image size is approximately 40×40×40 voxels, with each voxel containing complex impedance data at 20 frequency points. The resulting 3D image volume is approximately 20 MB and is used for further structural analysis and trend prediction. This image provides a complete "impedance autopsy map" of the capacitor in the spectral dimension. Based on multi-spectral impedance image data acquired at multiple moments and combined with the aforementioned tomographic results, the degradation trend of the internal structure of aluminum electrolytic capacitors over time is predicted, resulting in a time-sensitive health assessment indicator and graphical output. To achieve this prediction, a machine learning model is combined with comparative analysis of time-series images. First, multi-spectral 3D tomographic images of capacitor samples at different life stages (e.g., samples with usage times of 0, 200, 500, and 1000 hours) are acquired under the same test conditions. The multi-time images are aligned using a spatial registration algorithm (using rigid registration and ICP optimization) to ensure consistent spatial locations. Based on this, the impedance change trajectory at each voxel location over time is modeled. The voxel impedance evolution curve Z(t, f) is defined, and its temporal evolution trend parameters are extracted using nonlinear curve fitting methods (such as exponential decay models or polynomial fitting). A long short-term memory (LSTM) model is then used to perform time-series predictions of the impedance evolution trend at each location, outputting an impedance image for a specific moment in the future. The prediction results are reconstructed into a three-dimensional image, the "Impedance Tomography Degradation Trend Map." This image not only reflects the current structural state but also visualizes potential degradation areas and structural variation trends over time.

[0029] In this embodiment, the specific steps of predicting the structural degradation trend based on the multi-spectral impedance three-dimensional tomography image to construct the impedance tomography degradation trend prediction map are as follows: Identify electrolyte concentration distribution and extract electrolyte concentration interface based on multi-spectral impedance 3D tomography images; The impedance distribution difference analysis is performed on the multi-spectral impedance 3D tomography image to obtain the area with uneven oxide film thickness; Based on the electrolyte concentration interface and the uneven oxide film thickness area, point cloud modeling of abnormal areas is carried out, and spatial clustering analysis is performed to obtain a spatial clustering point cloud model of abnormal areas; Based on the spatial clustering point cloud model of the abnormal area, the volume proportion and distribution density are calculated, and the internal structure degradation is quantified to obtain the internal structure degradation quantitative index; Perform time-series impedance tomography comparative analysis on multi-spectral impedance 3D tomography images to extract the dynamic evolution trajectory of structural changes; The degradation trend of the dynamic evolution trajectory structure is predicted based on the quantitative index of internal structure degradation, thereby constructing an impedance tomography degradation trend prediction map.

[0030] In this embodiment, a fine impedance sampling device applies excitation signals at multiple frequency bands (e.g., 10kHz, 50kHz, 100kHz, 500kHz, and 1MHz) to the capacitor, constructing a five-band three-dimensional impedance tomography image. The image voxel size is 0.8mm × 0.8mm × 1mm, and each voxel reflects local variations in electrical impedance, corresponding to differences in electrolyte concentration and oxide film properties. Using a gradient field algorithm and impedance inversion reconstruction techniques, the system extracts impedance variations between voxels and identifies zones with sudden concentration gradients, representing electrolyte concentration interfaces, which characterize the heterogeneity of internal ion distribution. By calculating the differences between tomographic images across multiple frequency bands, a spectral response difference matrix is ​​constructed. This difference matrix is ​​used to analyze the differences in impedance response at the same spatial location under different excitation frequencies. Higher values ​​indicate the presence of regions with uneven oxide film thickness. The system uses a thresholding method (with a relative difference threshold of 12%) to identify these inconsistent regions and then performs voxel screening to generate an image of localized abnormal areas. After identifying the abnormal areas at the electrolyte interface and oxide film, the system overlays them, converting them into a three-dimensional coordinate dataset and constructing a point cloud model. Each point in the point cloud contains attribute information such as coordinates (x, y, z), impedance value, and frequency band identifier. To further analyze the spatial distribution characteristics of these abnormal points, a density clustering algorithm (such as DBSCAN, with a minimum sample size of 8 and an ε of 1.2 mm) is used to spatially cluster the point cloud data. This generates a spatial cluster point cloud model of the abnormal areas and identifies multiple high-density abnormal clusters. Based on the cluster model, the system calculates the volume ratio (cluster volume / total volume) and point cloud density (number of points per unit volume) of each cluster to assess the scope and extent of the degraded area within the capacitor structure. These two indicators are combined to form the Structural Degradation Quantification Index (SDQI), which quantifies the internal degradation state of the capacitor. A cluster volume ratio exceeding 8% or a density exceeding 100 points / mm³ is considered moderate or above degradation. The system incorporates the time dimension, performing time-series tomographic image registration on previously recorded impedance images at multiple stages. By using key feature point registration and a structural consistency algorithm, impedance images at multiple times are compared to extract the voxel evolution path of the abnormal region. This process constructs a dynamic evolution trajectory diagram, demonstrating the expansion rate, migration direction, and impedance trend of the degraded region. For example, in a sample test, a cluster of degraded points expanded by 2.4 mm³ within 24 hours, while its impedance value increased by 9.3%. Combining the aforementioned degradation quantification index (SDQI) with the dynamic evolution trajectory, the system constructs a degradation trend prediction model based on an LSTM (Long Short-Term Memory) network.The input is the evolution parameters extracted from historical time-series impedance images, and the output is a forecast of the spatial expansion trend and intensity of the degradation area in future time periods (e.g., 48 hours or 72 hours). The prediction results are presented as a three-dimensional heat map, called an impedance tomography degradation trend prediction map, which provides early warning of potential failure risks.

[0031] In this embodiment, refer to Figure 3 , is a flowchart of the detailed implementation steps of step S2. In this embodiment, the detailed implementation steps of step S2 include: The audio signal of the capacitor charging and discharging process is collected based on the audio sensor array; Decomposing the audio signal in the time-frequency domain to obtain a multi-scale time-frequency spectrum; Calculate the gray-level co-occurrence matrix of the multi-scale time-frequency spectrum, extract the contrast, correlation, energy and entropy values, and obtain the texture feature parameters; Perform statistical analysis on the amplitude distribution of texture feature parameters to identify the acoustic texture image of the capacitor; Long-term monitoring and trend analysis of the acoustic texture image of the capacitor are carried out, and the acoustic texture-structure degradation correlation mining is carried out based on the impedance tomography degradation trend prediction diagram to construct an acoustic texture damage evolution assessment model.

[0032] In this embodiment, a highly sensitive, high-resolution acoustic acquisition system is constructed using a multi-channel audio sensor array (e.g., a MEMS microphone array with 8 or 16 channels) arranged around an aluminum electrolytic capacitor. The system sampling rate is set to 44.1 kHz to ensure coverage of the human audible range and part of the ultrasonic band (high frequencies reflect subtle mechanical changes). The sampling time is typically 5 to 10 seconds per sample, covering a complete charge-rest-discharge cycle. The sensors are arranged symmetrically to ensure consistent spatial signal capture. They are positioned 3 to 5 cm from the capacitor surface to avoid contact interference while maintaining sufficient acoustic pressure sensitivity. After acquiring the raw audio signal, it must be converted into a more physically meaningful representation to capture the dynamic characteristics of the acoustic response. Therefore, the continuous wavelet transform (CWT) is selected as the primary tool for time-frequency analysis. Compared to the traditional short-time Fourier transform (STFT), the wavelet transform has higher time and frequency localization capabilities and is suitable for processing non-stationary signals, such as the spurious acoustic waves and harmonic responses generated by capacitor charging and discharging. Using the Morlet wavelet as the mother wavelet function, a multiscale decomposition was constructed with scaling factors ranging from 1 to 64. After wavelet transforming each channel of audio data, a three-dimensional time-frequency spectrum (time × frequency × channel) was obtained. To enhance the physical readability of the spectrum, the frequency range was limited to 100 Hz to 15 kHz, covering most capacitor structural vibration frequencies and the acoustic frequency bands potentially generated by partial discharge.

[0033] To facilitate image processing and subsequent texture feature extraction, the time-frequency spectrum is converted to two-dimensional data. Maximum projection and mean filtering are used to extract representative two-dimensional spectra (e.g., joint frequency and time spectra) from the three-dimensional data. These images are then resized to 128×128 pixels to facilitate standardization for subsequent texture analysis. The final output is one or more multi-scale time-frequency spectrograms corresponding to each experimental sample. The time-frequency spectrum obtained in the previous step is treated as a grayscale image, and its texture features are extracted using the Gray Level Co-occurrence Matrix (GLCM). The GLCM is a statistical method used to characterize the spatial distribution of image texture. Its core concept is to calculate the probability of occurrence of grayscale combinations within a certain direction and distance, thereby quantifying the image's structural information. The grayscale level is set to 64 (to quantize the 128×128 image), and four directions (0°, 45°, 90°, and 135°) and two distance scales (d=1 and 2) are selected for the calculation. A corresponding GLCM matrix is ​​generated for each image, and four typical texture parameters are extracted: contrast, correlation, energy, and entropy, representing the intensity of local variation, pixel correlation, image regularity, and information complexity, respectively. High contrast indicates that the energy in the time-frequency image fluctuates dramatically across the spectrum, potentially reflecting high-frequency transients caused by structural anomalies. High energy indicates that the overall grayscale of the image is concentrated, potentially corresponding to low noise in a stable structure. High entropy indicates that the acoustic signal is complex and fluctuates frequently, potentially related to capacitor material aging. The above method extracts a total of 32-dimensional texture features (4 directions × 2 distances × 4 features) from each image. These features serve as the structural representation vector for each acoustic sample, providing quantitative metrics for subsequent statistical distribution analysis and state discrimination. By statistically modeling the amplitudes of these texture features, the distribution of capacitors in different states in the acoustic texture space is characterized, thereby enabling the identification of "acoustic profiles." Typical statistical feature analysis methods, such as histogram distribution analysis, probability density estimation (KDE), and principal component analysis (PCA), were selected to model the texture feature dataset. The texture feature data of multiple capacitors with the same state (e.g., new, slightly aged, moderately aged, and severely aged) were clustered to generate a distribution map in a multidimensional feature space. PCA dimensionality reduction to two- or three-dimensional visualizations allows for intuitive observation of the clustering of samples in each state in the feature space. Capacitors with different aging states exhibit distinctly different distributions of texture parameter amplitudes. For example, severely aged capacitors typically have higher texture entropy and greater contrast. Unsupervised clustering methods, such as K-means and DBSCAN, were used to cluster the texture features to further verify their ability to distinguish capacitor states. Clustering quality was evaluated using metrics such as the Silhouette coefficient.Ultimately, an acoustic texture profile model is developed to identify the acoustic "fingerprint" of capacitors. This process can be understood as constructing an acoustic state template for the capacitor. Acoustic texture profiles of the capacitors are periodically collected (e.g., every 24 hours for a month) during long-term service, recording changes in texture features at each time point. Simultaneously, electrochemical impedance spectroscopy (EIS) is used to obtain impedance spectral parameters at corresponding time points. Characteristic indicators such as equivalent series resistance (ESR), capacitance (C), and phase angle are calculated to create a capacitor impedance tomography map. After acquiring these two types of data, time series correlation analysis methods (such as Spearman rank correlation and Granger causality tests) are used to explore the statistical correlation between changes in texture features and impedance. Particular attention is paid to the correlation between texture entropy and capacitance loss angle, as this often reflects non-uniform aging within the material plate. To further predict the degradation path of the capacitor, an acoustic texture damage evolution model is constructed. Machine learning regression methods (such as LSTM neural networks or random forest regression) are used, taking the texture feature sequence as input and the impedance spectral indicators as output labels, to predict degradation trends and estimate remaining useful life (RUL). The training data of this model comes from the periodic monitoring data accumulated in the early stage. The input features include texture contrast, entropy value and its rate of change. The model output is the prediction of the capacitor damage status in the next N days.

[0034] In this embodiment, reference Figure 4 The above is a flowchart of the detailed implementation steps of step S3. In this embodiment, the detailed implementation steps of step S3 include: During the multi-band excitation process, the integrated thermal infrared imaging device performs real-time thermal imaging monitoring of the capacitor surface to generate a thermal imaging image of the capacitor; Analyze the spatiotemporal evolution of temperature distribution on capacitor thermal imaging images and construct the spatiotemporal temperature distribution field; The temperature characteristics of different wavelengths are analyzed for the spatiotemporal temperature distribution field to obtain the spectral characteristic matrix of the temperature field; Constructing a heat conduction finite element simulation model according to the spectral characteristic matrix; Inversely infer the internal heat source distribution of the heat conduction finite element simulation model and locate the internal heat source distribution network; Abnormal heat flux density analysis is performed on the internal heat source distribution network, local overheating diffusion mining is carried out, and a three-dimensional thermal anomaly detection model is constructed.

[0035] In this example, to dynamically perceive the thermal characteristics of aluminum electrolytic capacitors, a high-sensitivity thermal infrared imager was used to capture real-time images of the surface temperature field during controlled excitation (e.g., AC pulses or frequency sweeps) of the capacitors. The excitation source was provided by a function signal generator with a frequency range of 100 Hz to 100 kHz and a voltage amplitude between 1 Vpp and 10 Vpp. This was intended to induce dielectric polarization, charge migration, and dielectric loss reactions within the capacitor's internal structure at different frequencies, thereby stimulating differences in thermal response. The thermal infrared imaging device used an uncooled thermal imager (such as the FLIR A655sc) with a sensitivity exceeding 50 mK and a spatial resolution of 640×480 pixels. This, combined with a high frame rate (30 Hz) and a macro lens, ensured accurate capture of minute surface temperature rises and hotspot diffusion. The camera was kept within 15 cm of the capacitor in an insulated test chamber to avoid environmental interference. Each excitation experiment lasted approximately 60 seconds, during which thermal images were continuously acquired and the excitation parameters and capacitor response signals were simultaneously recorded. Thermal imaging data is stored as a video or image sequence, with each frame accompanied by temperature calibration information (unit: °C). This data set forms the temperature evolution of the capacitor under multi-frequency excitation, providing a foundational map for subsequent thermal field analysis and modeling. A thermal image is a two-dimensional grayscale representation of the temperature field, depicting the surface temperature distribution of the capacitor at a specific moment. To characterize the dynamic thermal response of the capacitor throughout the entire excitation cycle, it is necessary to process it into a three-dimensional spatiotemporal temperature distribution field (spatial coordinates x and y + time t). This distribution field uses image frames as a time series, analyzing the temperature changes of each pixel frame by frame to form a complete data matrix of temperature variations over time. A bidirectional sliding window method is used to perform temporal filtering on the image sequence to remove abnormal fluctuations introduced by environmental jitter or optical errors. In the spatial dimension, to more precisely identify hotspots, a two-dimensional Gaussian filter and edge enhancement algorithm are used to extract areas with prominent temperature changes. Temperature processing uses metrics such as the absolute temperature gradient (ΔT / Δt) and the growth rate of local hot spots to characterize the dynamic evolution process.

[0036] Multiple regions of interest (ROIs) were set for different parts of the capacitor (leads, casing, electrode tops), and their respective temperature evolution curves were extracted and fitted with trend lines. Generally, a healthy capacitor's thermal response should exhibit a rapid temperature increase followed by stabilization. However, if internal defects, leakage, or electrochemical aging are present, the temperature evolution curve will often exhibit nonlinear growth or delayed hot spots in certain areas, indicating local structural abnormalities. Although thermal infrared images are essentially visible grayscale temperature projections, in practice, different wavelengths of infrared radiation have different penetration and contrast, revealing thermal characteristics at different depths within a material or at the surface. Therefore, in this step, if the thermal imaging equipment supports multi-band imaging (such as LWIR 8–14μm and MWIR 3–5μm), the same thermal field can be simultaneously imaged at multiple wavelengths. Thermal images acquired at these multiple bands are registered (using feature point matching and image reprojection algorithms) to establish a one-to-one correspondence in spatial and temporal dimensions, thereby constructing a four-dimensional temperature matrix (wavelength λ × x × y × t). Based on this, wavelength response curves are extracted at typical moments (e.g., peak temperature rise points). The spectral components of the temperature response are analyzed using Fourier transforms (FFTs) to generate temperature spectra for different regions. Capacitors with different aging states or structural defects exhibit different radiation contrast in the mid-infrared and far-infrared bands. For example, moderately aged capacitors exhibit higher hot spot contrast in the MWIR band, indicating that the heat source is close to the material surface; while severely aged capacitors exhibit a continuously expanding hot zone in the LWIR band, indicating limited heat conduction. After obtaining detailed spatiotemporal temperature evolution data and spectral response parameters, these experimental data are used to construct a finite element simulation (FEM) model based on heat conduction theory. This model aims to infer the distribution of heat sources within the internal structure from a heat flow perspective, thereby enabling non-contact diagnosis of the capacitor's internal condition. A 3D model (based on COMSOL Multiphysics or ANSYS Workbench) consistent with the actual capacitor geometry is constructed, including the anode aluminum foil, electrolyte, insulation sleeve, and lead structure. Material parameters such as thermal conductivity, specific heat capacity, and density are input according to literature or actual test values. For example, the thermal conductivity of aluminum foil is 237 W / m·K, and that of the electrolyte is 0.6 W / m·K. The simulation boundary conditions are set according to the experimental thermal image. The surface temperature uses the measured temperature value in the spectral matrix as input. Virtual heat source points or heat source areas are introduced into the model and initially set to multiple possible positions. The heat conduction simulation adopts two modes: steady-state and transient. The time step is set to 0.1 seconds, and the total simulation time is consistent with the experimental acquisition time. By adjusting the spatial position and power distribution of the internal heat source and continuously iterating the comparison between the simulation results and the actual thermal map until the error between the two (such as the maximum temperature difference error <1.5°C) is minimized, the construction of the heat conduction inversion model is finally completed, providing a basis for the subsequent positioning of the internal heat source.Finite element models are used to perform inverse analysis on experimentally observed thermal imaging data to identify the hidden heat source distribution network within the capacitor. Optimization algorithms (such as genetic algorithms (GA) or particle swarm optimization (PSO)) are then used to inversely optimize the internal heat source parameters of the simulation model to match the external thermal image observations.

[0037] Multiple potential heat source areas are defined, such as the anode folding layer, residual electrolyte area, and electrode contact points. By controlling the heat flux density (unit: W / m²) and distribution of each heat source, the surface temperature field output by the model is gradually adjusted. The results are then compared with the measured thermal images to evaluate the error distribution (using the RMSE or SSIM metrics).

[0038] The heat source of new capacitors is primarily concentrated in the contact area between the pins and the aluminum foil; however, the heat source of degraded capacitors is distributed discretely at multiple points, exhibiting typical characteristics of electrolyte evaporation and leakage hotspots. The final output is an internal heat source distribution map, annotating the spatial location and heat intensity of each heat source, forming a complete heat source distribution network, providing a structural basis for further fault identification. After obtaining the three-dimensional distribution network of internal heat sources in the capacitor, the final step focuses on identifying abnormal heat flux and modeling thermal diffusion behavior. Abnormally high heat flux density often indicates internal structural degradation, material aging, or electrochemical reaction anomalies. Gradient analysis of the heat flux density distribution output by the simulation identifies localized heat flux abrupt changes. The presence of abnormal heat diffusion paths is determined by combining the spatial gradient of the heat diffusion rate (the spatial gradient of the derivative ΔT / Δt). Criteria for identifying abnormal areas include: a local heat flux density exceeding three times the average; a continuous expansion of the hot spot area exceeding a 20% inter-frame growth rate; or inconsistent temperature responses of the same heat source area across different wavelengths. Using these thermal anomaly characteristics, a three-dimensional thermal anomaly detection model is constructed. Each abnormal region is encoded in the spatial (x, y, z) and temporal (t) dimensions to form a multidimensional anomaly indicator vector. This model can be trained for automatic thermal anomaly identification and alarming, and can also serve as input for capacitor thermal failure prediction.

[0039] In this embodiment, step S4 includes the following steps: Perform in-situ cyclic voltammetry on the capacitor electrolyte to obtain the current-voltage characteristic curve of the electrochemical reaction; Analyze the shape change of the curve and the current peak attenuation trend according to the current-voltage characteristic curve; The diffusion coefficient of the active material in the electrolyte is obtained by quantitatively evaluating the diffusion of the active material based on the shape change and the current peak decay trend. Obtain electrolyte pH value and subconcentration chromatogram during in-situ cyclic voltammetry testing; Track the changes in electrolyte chemical composition based on electrolyte pH and sub-concentration chromatograms to generate chemical composition change trajectories; The electrolyte degradation trend is analyzed based on the diffusion coefficient of active substances and the change trajectory of chemical components, and the electrolyte degradation trend curve is constructed.

[0040] This example quantitatively analyzes the dynamic changes in the electrochemical properties and chemical composition of the electrolyte within an aluminum electrolytic capacitor under operating conditions. In situ cyclic voltammetry (CV) testing was used to obtain the electrolyte's current-voltage characteristic curve. Combined with pH and chromatographic analysis data, a model for electrolyte degradation was constructed to aid in capacitor life prediction and condition assessment. In situ CV testing of the electrolyte was performed. This experiment employed a three-electrode system, with an aluminum electrode as the working electrode, a standard hydrogen electrode as the reference electrode, and a platinum electrode as the counter electrode. The test voltage sweep range was set between -0.2 V and 1.2 V, with a scan rate of 50 mV / s, and repeated at least 50 times to ensure data stability. During the test, the current response was recorded as the voltage varied, forming a detailed current-voltage (IV) characteristic curve. This curve reflects the redox reactions within the electrolyte, electrode interface processes, and charge transfer dynamics. The IV curve was analyzed for morphological changes and peak decay trends. By comparing the curves from the initial cycle with those from subsequent cycles, changes in current intensity, peak shift, and curve area of ​​the oxidation and reduction peaks were observed. Generally speaking, the gradual decay of the current peak reflects the gradual consumption of active materials or interfacial passivation, while peak shifts indicate changes in reaction kinetics. For quantitative analysis, peak area integration and peak height calculation, combined with baseline correction, were used to accurately extract peak current time series data. In the experiment, the peak current gradually decayed from an initial 2.5 mA to 1.6 mA, with a decay rate of approximately 36%, reflecting the gradual depletion of the electrolyte's active components. Based on these curves, the diffusion coefficients of the electrolyte's active materials were quantitatively assessed. The Randles-Sevcik equation was used to estimate the diffusion coefficients, D, of the active materials in the electrolyte based on the relationship between peak current and scan rate. Specifically, a linear fit was performed on the peak current data at different scan rates (10 mV / s to 100 mV / s), with the slope directly related to the diffusion coefficient. The experimental results showed that D decreased from 1.2×10-6 cm² / s to 7.8×10-7 cm² / s with electrolyte aging, reflecting a decrease in the diffusion capacity of the active materials and indicative of electrolyte performance degradation. During the test, electrolyte pH and sub-concentration chromatography data were collected to track changes in chemical components. An online pH sensor with an accuracy of ±0.01 and a monitoring range of 3 to 7 was used for pH measurement. During the experiment, the electrolyte pH slowly rose from an initial pH of 4.5 to 5.3, indicating a change in the acid-base environment. Sub-concentration chromatography analysis quantified the main electrolyte components (such as sulfate and aluminum ions) using an ion chromatograph. The data collection cycle was 1 hour, and the detection sensitivity reached the ppb level. The results revealed that the concentrations of the main active components changed slightly but steadily over time, with sulfate ion concentrations decreasing by 12% and aluminum ion concentrations showing a slight accumulation. Combining the diffusion coefficients of active substances with changes in chemical composition, a systematic electrolyte degradation trend analysis model was constructed.The model, based on multivariate time series analysis, generates a comprehensive degradation profile by weightedly integrating the decreasing trend of the diffusion coefficient, the pH curve, and ion concentration fluctuations. This profile, with time as the horizontal axis and the degradation index as the vertical axis, vividly depicts the dynamic process of electrolyte performance degradation. In experiments, the degradation profile clearly exhibits an exponential decay trend. Validated with actual capacitor lifespan data, the model's prediction error is kept within 5%.

[0041] In this embodiment, the specific steps of step S5 are: Based on the acoustic texture damage evolution assessment model, the three-dimensional thermal anomaly detection model, and the electrolyte degradation situation curve, multi-dimensional feature space mapping is performed to construct the three-dimensional state feature space of the capacitor; Standardizing and normalizing the three-dimensional state feature space of the capacitor to obtain a standardized three-dimensional state feature space; Fuzzy logic reasoning is performed between multi-dimensional features in the standardized three-dimensional state feature space to evaluate uncertainty and conflict information, so as to obtain fuzzy evaluation results between multi-dimensional features.

[0042] In this embodiment, multi-source monitoring data is integrated, fusing acoustic texture damage evolution, three-dimensional thermal anomaly detection, and electrolyte degradation status information to construct a three-dimensional feature space that comprehensively reflects the current health status of aluminum electrolytic capacitors. Fuzzy logic methods are then used to perform correlation analysis and uncertainty processing of multi-dimensional features, providing a scientific basis for accurate condition assessment. Based on three key feature models obtained previously—the acoustic texture damage evolution assessment model, the three-dimensional thermal anomaly detection model, and the electrolyte degradation status curve—corresponding state parameter data are extracted at a unified time point. The acoustic texture damage assessment model primarily outputs quantitative indicators of microcracks, spalling, and structural anomalies within the capacitor, with values ​​typically ranging from 0 to 1, indicating the degree of damage. The three-dimensional thermal anomaly detection model uses infrared thermal imaging and temperature gradient calculations to determine the size and severity of the thermal anomaly area, expressed as a degree Celsius difference and area percentage. The electrolyte degradation status curve provides a comprehensive degradation index for electrolyte degradation, typically presented as a normalized numerical value. These three parameters are used as the three dimensional coordinates of the three-dimensional feature space to construct a three-dimensional state feature space for the capacitor, which comprehensively displays the complex state of the capacitor's internal structure and chemical properties. To eliminate the impact of dimensional differences and the value spans of different indicators, the constructed three-dimensional feature space is standardized and normalized. Standardization typically uses the Z-score method, which subtracts the mean from each dimension and divides by the standard deviation. This centralizes the data distribution and provides a uniform standard deviation, facilitating subsequent statistical analysis. Normalization uses linear scaling to linearly transform the data to the [0, 1] interval, ensuring that the weights of each dimensional indicator are balanced within the same numerical range. In the experiment, the acoustic indicator (mean 0.35, standard deviation 0.12), the thermal anomaly indicator (mean 5.2°C, standard deviation 1.8°C), and the degradation index (mean 0.48, standard deviation 0.15) were standardized and then uniformly normalized to the range of 0 to 1. This process effectively prevents evaluation bias caused by excessively large or small values ​​in a single dimension.

[0043] Because capacitor status information is subject to uncertainty due to multi-source data, sensor errors, and environmental influences, a fuzzy logic system can effectively handle ambiguity and conflict, avoiding the bias of traditional hard threshold judgments. First, based on the physical and chemical properties of the capacitor status, fuzzy sets and their membership functions are defined within a three-dimensional feature space. For example, acoustic damage severity is categorized as "low," "medium," and "high," thermal anomalies are categorized as "normal," "warning," and "severe," and degradation indices are categorized as "good," "medium," and "degraded." Each category corresponds to a membership function with a different shape (such as triangular or Gaussian), and the membership function parameters are determined through expert experience and historical experimental data. A fuzzy rule base is established, with rules such as "If the acoustic damage is high, the thermal anomaly is severe, and the degradation index is degraded, then the overall status is severely degraded." The number of rules is typically set between 20 and 30 to cover typical status combinations. During the experiment, expert review and historical data verification ensure the scientific nature and comprehensiveness of the rules. Based on this, the system uses normalized three-dimensional feature data as input and performs inference calculations using fuzzy inference mechanisms (such as the Mamdani inference model) to assess conflicts and uncertainties among multi-dimensional features and obtain a fuzzy assessment output. Through the defuzzification process of the fuzzy output, the centroid method is used to obtain a single numerical assessment result, representing the overall health status of the capacitor at that moment. This result reflects the complex relationships between multi-dimensional data, improving the robustness and accuracy of the condition assessment. Experimental results show that this fuzzy logic assessment method reduces the error by approximately 15% compared to traditional linear weighted models. In multiple real-world tests, the response time has been kept within 50ms, meeting the requirements of online monitoring.

[0044] In this embodiment, the specific steps of step S6 are: Obtaining historical capacitor failure cases and performing incremental learning optimization on the fuzzy evaluation results to obtain learning optimization samples of multiple evaluation results; Performing multi-level health status discrimination on the learning optimization samples to construct a hierarchical classification system for failure modes; Conduct capacitor failure risk level assessment based on the failure mode hierarchical classification system to obtain a failure risk level assessment report; Capacitor performance degradation prediction is performed based on the acoustic texture damage evolution assessment model and the electrolyte degradation trend curve, and a performance degradation assessment report is generated; The final capacitor status assessment result is obtained based on the fault risk level assessment report and the performance degradation assessment report.

[0045] In this embodiment, a fuzzy logic assessment model is optimized through incremental learning, drawing on a large number of historical capacitor failure cases to improve the accuracy and adaptability of condition assessment. The optimization results are then used to implement multi-level health state discrimination and establish a hierarchical failure mode classification system. This is followed by fault risk level assessment and performance degradation prediction, ultimately forming a comprehensive capacitor condition assessment system. Historical capacitor failure case data is obtained. These cases typically number hundreds to thousands of instances, including collected signals, fault diagnosis conclusions, and related environmental parameters for various types of actual operation. The fuzzy logic assessment results obtained in the previous steps are compared and analyzed with this historical failure data. Incremental learning is then used to continuously optimize the fuzzy assessment model, combining the labeled fault categories and health states. Incremental learning employs online training algorithms such as support vector machines (SVMs) or random forests (RFs), which dynamically adjust model parameters as new data arrives, avoiding the high cost of retraining and data forgetting. Experimental parameters were set: an incremental learning batch size of 50 samples and a gradually decreasing learning rate. Training accuracy exceeded 92% through cross-validation. This method constructs a fuzzy evaluation optimization sample set at multiple time points and under different operating conditions, significantly improving the model's adaptability and generalization capabilities. Multi-level health status discrimination is performed on the learned optimization samples. By analyzing the optimized evaluation results, a hierarchical clustering algorithm (such as hierarchical clustering combined with K-means) is used to divide the capacitor status into multiple health levels, generally divided into four levels: "healthy," "mildly degraded," "moderately degraded," and "severely degraded." The fuzzy evaluation value interval corresponding to each level is obtained through statistical analysis of historical data. For example, the healthy interval is [0, 0.3], mildly degraded is (0.3, 0.5], moderately degraded is (0.5, 0.7], and severely degraded is (0.7, 1.0). At the same time, a hierarchical classification system is constructed based on the similarity of fault modes, subdividing the fault modes into electrolyte degradation, structural damage, thermal anomaly, etc., forming a multi-level fault discrimination framework that supports multi-angle and multi-level health status diagnosis.

[0046] Based on this hierarchical classification system of fault modes, a capacitor fault risk level assessment was carried out. The risk assessment combines the severity, probability of occurrence, and potential impact of the fault, and a weighted scoring method is used to calculate the risk level. The weight allocation refers to expert experience and historical statistical data, such as a severity weight of 0.5, a probability of occurrence of 0.3, and an impact range of 0.2. The calculated results correspond to four risk levels: low, medium, high, and very high. Dynamic risk monitoring is achieved by comparing the temporal changes in risk levels. In experimental verification, the risk assessment accuracy reached 88%, and the average advance warning time for predicting critical faults was 72 hours, which has good practical value. The assessment results are compiled to generate a fault risk level assessment report, covering risk levels, fault type distribution, and historical trends.

[0047] Capacitor performance degradation is predicted based on an acoustic texture damage evolution assessment model and electrolyte degradation curves. A multivariate time series prediction model (such as a long short-term memory (LSTM) network) is established, fed with historical acoustic and electrochemical degradation indicators, to predict performance indicator trends over a period of time. The model is trained using a 30-day historical data window and a 7-day prediction step, achieving a prediction accuracy of less than 5% mean square error. The output is presented as a performance degradation curve, visually reflecting the remaining life trend and performance degradation rate of the capacitor. This generates a performance degradation assessment report, providing a scientific basis for maintenance decisions. A comprehensive analysis is conducted by combining the fault risk level assessment report with the performance degradation assessment report to produce the final capacitor condition assessment result. This comprehensive assessment result not only reflects the current health status and risk level but also includes a forecast of future performance trends, forming a complete closed-loop condition monitoring system. Dynamic data updates ensure real-time and accurate assessment results, supporting intelligent early warning and maintenance strategy adjustments. The overall system has demonstrated high stability in multiple field applications, effectively reducing unplanned downtime and maintenance costs.

[0048] In this embodiment, a device for evaluating the state of an aluminum electrolytic capacitor based on multi-dimensional data analysis is provided, which is used to perform the above-mentioned method for evaluating the state of an aluminum electrolytic capacitor based on multi-dimensional data analysis, including: The impedance tomography module is used to perform multi-band excitation on aluminum electrolytic capacitors, reconstruct three-dimensional fault distribution, and predict structural degradation trends, thereby constructing an impedance tomography degradation trend prediction map; The acoustic texture module is used to collect audio signals from the capacitor's charging and discharging processes, perform acoustic texture analysis, and mine the soundprint-structure degradation correlation based on the impedance tomography degradation trend prediction chart to build an acoustic texture damage evolution assessment model. The temperature distribution module is used to obtain thermal imaging images of capacitors, analyze the spatiotemporal evolution of temperature distribution, and conduct local overheating diffusion mining to build a three-dimensional thermal anomaly detection model; Degradation trend analysis module, used to track changes in electrolyte chemical composition and analyze electrolyte degradation trends of capacitor electrolytes, and construct electrolyte degradation trend curves; The feature space mapping module is used to perform multi-dimensional feature space mapping and fuzzy logic reasoning between features based on the acoustic texture damage evolution assessment model, the three-dimensional thermal anomaly detection model, and the electrolyte degradation situation curve to obtain fuzzy evaluation results between multi-dimensional features; The comprehensive status assessment module is used to assess the capacitor failure risk level according to the fuzzy assessment result and predict the capacitor performance attenuation to perform the capacitor status assessment operation.

[0049] The present invention is therefore intended to be illustrative and non-restrictive in all respects, with the scope of the invention being defined by the appended claims rather than the foregoing description, and all changes that come within the meaning and range of equivalents of the application documents are intended to be embraced therein.

[0050] The foregoing description is intended only to provide specific embodiments of the present invention, which will enable those skilled in the art to understand and implement the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not intended to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for evaluating the status of aluminum electrolytic capacitors based on multi-dimensional data analysis, characterized in that: The following steps are involved: Step S1: performing multi-band excitation on the aluminum electrolytic capacitor, and performing three-dimensional fault distribution reconstruction and structural degradation trend prediction, thereby constructing an impedance tomography degradation trend prediction map; Step S2: Collect audio signals during the capacitor's charging and discharging process, perform acoustic texture analysis, and mine the soundprint-structure degradation correlation based on the impedance tomography degradation trend prediction diagram to construct an acoustic texture damage evolution assessment model. Step S3: Acquire thermal imaging images of the capacitor, perform spatiotemporal evolution analysis of temperature distribution, conduct local overheating diffusion mining, and construct a three-dimensional thermal anomaly detection model; Step S4: Tracking the chemical composition changes of the capacitor electrolyte and analyzing the electrolyte degradation trend, and constructing an electrolyte degradation trend curve; Step S5: Based on the acoustic texture damage evolution assessment model, the three-dimensional thermal anomaly detection model, and the electrolyte degradation situation curve, multi-dimensional feature space mapping and fuzzy logic reasoning between features are performed to obtain fuzzy assessment results between multi-dimensional features; Step S6: performing capacitor failure risk level assessment based on the fuzzy assessment result and capacitor performance degradation prediction to perform capacitor status assessment.

2. The aluminum electrolytic capacitor status assessment method based on multi-dimensional data analysis according to claim 1, characterized in that: The specific steps of step S1 are: Define a multi-band impedance measurement excitation source, apply multi-band excitation to aluminum electrolytic capacitors, perform omnidirectional impedance synchronous scanning based on a ring electrode array, and extract omnidirectional multi-band capacitor excitation response parameters; Calculating impedance values ​​of the capacitance excitation response parameters in different directions, performing spatial distribution fitting, and constructing an impedance distribution space coordinate mapping network; The three-dimensional fault distribution is reconstructed based on the impedance distribution space coordinate mapping network to construct a multi-spectral impedance three-dimensional tomographic image; The structural degradation trend is predicted based on the multi-spectral impedance three-dimensional tomography images, thereby constructing the impedance tomography degradation trend prediction map.

3. The aluminum electrolytic capacitor status assessment method based on multi-dimensional data analysis according to claim 2, characterized in that: The specific steps of predicting the structural degradation trend based on the multi-spectral impedance three-dimensional tomography image and thus constructing the impedance tomography degradation trend prediction map are as follows: Identify electrolyte concentration distribution and extract electrolyte concentration interface based on multi-spectral impedance 3D tomography images; The impedance distribution difference analysis is performed on the multi-spectral impedance 3D tomography image to obtain the area with uneven oxide film thickness; Based on the electrolyte concentration interface and the uneven oxide film thickness area, point cloud modeling of abnormal areas is carried out, and spatial clustering analysis is performed to obtain a spatial clustering point cloud model of abnormal areas; Based on the spatial clustering point cloud model of the abnormal area, the volume proportion and distribution density are calculated, and the internal structure degradation is quantified to obtain the internal structure degradation quantitative index; Perform time-series impedance tomography comparative analysis on multi-spectral impedance 3D tomography images to extract the dynamic evolution trajectory of structural changes; The degradation trend of the dynamic evolution trajectory structure is predicted based on the quantitative index of internal structure degradation, thereby constructing an impedance tomography degradation trend prediction map.

4. The aluminum electrolytic capacitor status assessment method based on multi-dimensional data analysis according to claim 1, characterized in that: The specific steps of step S2 are: The audio signal of the capacitor charging and discharging process is collected based on the audio sensor array; Decomposing the audio signal in the time-frequency domain to obtain a multi-scale time-frequency spectrum; Calculate the gray-level co-occurrence matrix of the multi-scale time-frequency spectrum, extract the contrast, correlation, energy and entropy values, and obtain the texture feature parameters; Perform statistical analysis on the amplitude distribution of texture feature parameters to identify the acoustic texture image of the capacitor; Long-term monitoring and trend analysis of the acoustic texture image of the capacitor are carried out, and the acoustic texture-structure degradation correlation mining is carried out based on the impedance tomography degradation trend prediction diagram to construct an acoustic texture damage evolution assessment model.

5. The aluminum electrolytic capacitor status assessment method based on multi-dimensional data analysis according to claim 1, characterized in that: The specific steps of step S3 are: During the multi-band excitation process, the integrated thermal infrared imaging device performs real-time thermal imaging monitoring of the capacitor surface to generate a thermal imaging image of the capacitor; Analyze the spatiotemporal evolution of temperature distribution on capacitor thermal imaging images and construct the spatiotemporal temperature distribution field; The temperature characteristics of different wavelengths are analyzed for the spatiotemporal temperature distribution field to obtain the spectral characteristic matrix of the temperature field; Constructing a heat conduction finite element simulation model according to the spectral characteristic matrix; Inversely infer the internal heat source distribution of the heat conduction finite element simulation model and locate the internal heat source distribution network; Abnormal heat flux density analysis is performed on the internal heat source distribution network, local overheating diffusion mining is carried out, and a three-dimensional thermal anomaly detection model is constructed.

6. The aluminum electrolytic capacitor status assessment method based on multi-dimensional data analysis according to claim 1, characterized in that: The specific steps of step S4 are: Perform in-situ cyclic voltammetry on the capacitor electrolyte to obtain the current-voltage characteristic curve of the electrochemical reaction; Analyze the shape change of the curve and the current peak attenuation trend according to the current-voltage characteristic curve; The diffusion coefficient of the active material in the electrolyte is obtained by quantitatively evaluating the diffusion of the active material based on the shape change and the current peak decay trend. Obtain electrolyte pH value and subconcentration chromatogram during in-situ cyclic voltammetry testing; Track the changes in electrolyte chemical composition based on electrolyte pH and sub-concentration chromatograms to generate chemical composition change trajectories; The electrolyte degradation trend is analyzed based on the diffusion coefficient of active substances and the change trajectory of chemical components, and the electrolyte degradation trend curve is constructed.

7. The aluminum electrolytic capacitor status assessment method based on multi-dimensional data analysis according to claim 1, characterized in that: The specific steps of step S5 are: Based on the acoustic texture damage evolution assessment model, the three-dimensional thermal anomaly detection model, and the electrolyte degradation situation curve, multi-dimensional feature space mapping is performed to construct the three-dimensional state feature space of the capacitor; Standardizing and normalizing the three-dimensional state feature space of the capacitor to obtain a standardized three-dimensional state feature space; Fuzzy logic reasoning is performed between multi-dimensional features in the standardized three-dimensional state feature space to evaluate uncertainty and conflict information, so as to obtain fuzzy evaluation results between multi-dimensional features.

8. The aluminum electrolytic capacitor status assessment method based on multi-dimensional data analysis according to claim 1, characterized in that: The specific steps of step S6 are: Obtaining historical capacitor failure cases and performing incremental learning optimization on the fuzzy evaluation results to obtain learning optimization samples of multiple evaluation results; Performing multi-level health status discrimination on the learning optimization samples to construct a hierarchical classification system for failure modes; Conduct capacitor failure risk level assessment based on the failure mode hierarchical classification system to obtain a failure risk level assessment report; Capacitor performance degradation prediction is performed based on the acoustic texture damage evolution assessment model and the electrolyte degradation trend curve, and a performance degradation assessment report is generated; The final capacitor status assessment result is obtained based on the fault risk level assessment report and the performance degradation assessment report.

9. A device for evaluating the status of an aluminum electrolytic capacitor based on multi-dimensional data analysis, characterized in that: The method for evaluating the state of an aluminum electrolytic capacitor based on multi-dimensional data analysis according to claim 1 comprises: The impedance tomography module is used to perform multi-band excitation on aluminum electrolytic capacitors, reconstruct three-dimensional fault distribution, and predict structural degradation trends, thereby constructing an impedance tomography degradation trend prediction map; The acoustic texture module is used to collect audio signals from the capacitor's charging and discharging processes, perform acoustic texture analysis, and mine the soundprint-structure degradation correlation based on the impedance tomography degradation trend prediction chart to build an acoustic texture damage evolution assessment model. The temperature distribution module is used to obtain thermal imaging images of capacitors, analyze the spatiotemporal evolution of temperature distribution, and conduct local overheating diffusion mining to build a three-dimensional thermal anomaly detection model; Degradation trend analysis module, used to track changes in electrolyte chemical composition and analyze electrolyte degradation trends of capacitor electrolytes, and construct electrolyte degradation trend curves; The feature space mapping module is used to perform multi-dimensional feature space mapping and fuzzy logic reasoning between features based on the acoustic texture damage evolution assessment model, the three-dimensional thermal anomaly detection model, and the electrolyte degradation situation curve to obtain fuzzy evaluation results between multi-dimensional features; The comprehensive status assessment module is used to assess the capacitor failure risk level according to the fuzzy assessment result and predict the capacitor performance attenuation to perform the capacitor status assessment operation.

Images