Metallized film capacitor initial self-healing type classification method

Abstract

The invention relates to the technical field of capacitor classification, and discloses a metallized film capacitor initial self-healing type classification method comprising the following steps: applying voltages with different boost rates to a to-be-tested metallized film capacitor, collecting key data when self-healing occurs, comprising images of self-healing voltage, power fluctuation data in a self-healing process, self-healing point temperature rise, a self-healing time range and a self-healing point fracture surface; based on the self-healing voltage and power fluctuation data in the self-healing process, constructing an energy dissipation model, and calculating an energy dissipation rate in unit time; and determining the maximum temperature rise of the self-healing point based on the energy dissipation rate and the heat conduction coefficient of the material. Through the technical scheme of combining multi-dimensional feature extraction with Gaussian mixture model classification, the technical effect of accurately classifying the initial self-healing type of the metallized film capacitor is achieved, and the problems that the classification result is insufficient in precision, difficult to quantify and poor in repeatability are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of capacitor classification, and in particular to a method for classifying the initial self-healing types of metallized film capacitors. Background Art

[0002] With the widespread use of electronic devices, metallized film capacitors (MTCs) are widely used in power, electronics, and industrial control due to their excellent performance. However, due to the various voltage fluctuations and thermal effects experienced by MTCs during operation, the randomness and complexity of their initial self-healing behavior pose significant challenges to capacitor performance evaluation and reliability prediction. Without a scientific classification method, the random nature of self-healing performance leads to low classification accuracy, difficulty in quantification, and unstable subsequent reliability assessment results. To this end, a variety of self-healing behavior evaluation and classification methods have been developed. Many of these methods attempt to construct preliminary classification models by analyzing the changes in the self-healing voltage and current of metallized film capacitors under different conditions. Existing methods for evaluating the self-healing performance of metallized film capacitors typically rely on simple electrical performance tests or on obtaining dynamic characteristics of the capacitors through online monitoring equipment. These methods primarily rely on collecting a single characteristic of the self-healing voltage or current and combining it with empirical judgment to classify the self-healing performance and evaluate the applicable voltage range. However, the existing classification method for metallized film capacitors is too simplistic and relies on a single feature such as self-healing voltage or current. It is difficult to fully capture the complex behavior in the self-healing process and cannot effectively solve the randomness problem of self-healing performance. In addition, for the spatial distribution characteristics of self-healing performance, there is a lack of in-depth analysis of the geometric complexity and spatial characteristics of the fracture surface, resulting in the classification results not being consistent with the actual performance. The experimental method is inefficient and costly, especially in the high-precision acquisition of dynamic signals, which shows obvious deficiencies and cannot meet the needs of large-scale production and quality inspection.

[0003] Therefore, the present invention proposes a method for classifying the initial self-healing types of metallized film capacitors to address the deficiencies of the prior art. Summary of the Invention

[0004] In response to the shortcomings of the existing technology, the present invention provides a method for classifying the initial self-healing types of metallized film capacitors, which solves the problem that the existing classification methods of metallized film capacitors are too single and rely on single characteristics such as self-healing voltage or current, making it difficult to fully capture the complex behavior in the self-healing process and effectively solve the randomness problem of self-healing performance. In addition, for the spatial distribution characteristics of the self-healing performance, there is a lack of in-depth analysis of the geometric complexity and spatial characteristics of the fracture surface, resulting in the classification results being inconsistent with the actual performance.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: A method for classifying the initial self-healing types of metallized film capacitors, comprising the following steps: Apply voltages at different ramp rates to the metallized film capacitor under test and collect key data during self-healing, including self-healing voltage, power fluctuation data during the self-healing process, temperature rise at the self-healing point, self-healing time range, and images of the fracture surface at the self-healing point. Based on the self-healing voltage and the power fluctuation data during the self-healing process, an energy dissipation model is constructed to calculate the energy dissipation rate per unit time; Based on the energy dissipation rate and the thermal conductivity of the material, the maximum temperature rise of the self-healing point is determined; a random process dynamics model is established for the change of the self-healing voltage over time to extract the dissipation coefficient and fluctuation intensity of the self-healing voltage change; Based on the fracture surface image of the self-healing point, the fractal dimension of the fracture surface is calculated and the fracture complexity characteristics of the self-healing point are extracted; The initial self-healing type of metallized film capacitors is classified using a classification model based on classification features.

[0006] Preferably, the voltage range of the different voltage increasing rates applied to the metallized film capacitor to be tested is 0.5 volts per second to 10 volts per second, the applied voltage range is zero to two thousand volts, and the collected data includes the following: The voltage value when self-healing occurs; Instantaneous voltage and current data during the self-healing process; Self-healing power fluctuation curve; Time series changes of temperature rise at the self-healing point; The start and end time of self-healing.

[0007] The data acquisition frequency is one million times per second, which can meet the demand for high-precision dynamic change capture. Preferably, the energy dissipation rate per unit time is calculated by the following formula: Where Φ represents the energy release rate per unit time during the self-healing process, t0 represents the time point when the self-healing process starts, P(t) represents the instantaneous power at time t, and t c Indicates the time point when the self-healing process ends, dt indicates the time interval [t0,t c ] is a very small time increment within .

[0008] Preferably, the maximum temperature rise of the self-healing point is determined by the following process: Obtain energy release data during the self-healing process, including energy dissipation per unit time; According to the thermal conductivity of the metallized film capacitor material, the energy released per unit time is converted into temperature rise; The temperature change of the self-healing point is collected in real time, and its maximum temperature rise value is recorded. The temperature rise value reflects the heat dissipation characteristics of the metallized film capacitor material during the self-healing process; Temperature rise data combined with material characteristic parameters are used as important feature quantities for subsequent classification models.

[0009] Preferably, the stochastic process dynamics model is described by the following stochastic differential equation: dU(t)=-αU(t)dt+σdW t Where U(t) is the time variation of the self-healing voltage, dU(t) represents the variation of the self-healing voltage U(t) within a small time interval of time t, α represents the decay rate of the voltage over time during the self-healing process, σ represents the intensity of random fluctuations in the self-healing voltage variation, αU(t) represents the deterministic dissipation term of the voltage, and dW t represents the increment of Brownian motion in a random process at time t.

[0010] Preferably, the stochastic process dynamics model extracts the dissipation coefficient and the fluctuation intensity by the following steps: the dissipation coefficient is calculated by the following formula: Among them, α represents the decay rate of voltage over time during the self-healing process, dU(t) represents the change value of the self-healing voltage U(t) within a small time interval of time t, Indicates the rate of change of voltage over time; The fluctuation intensity is extracted through statistical analysis of the self-healing voltage, specifically the standard deviation of the self-healing voltage; The data source includes the dynamic capture of the self-healing voltage by a high-frequency sampling system with a sampling frequency of one million times per second.

[0011] Preferably, the fractal dimension of the fracture surface image is calculated by the following formula: Among them, D f It represents the geometric complexity of the fracture surface image, N(r) is the number of covering units with a scale of r on the fracture surface, r represents the unit scale used to cover the fracture surface, log(N(r)) represents the logarithm of the number of units N(r) required to cover the fracture surface, and log(1 / r) represents the logarithm of the reciprocal of the measurement scale r.

[0012] Preferably, the classification features include the following five main features: Energy dissipation rate per unit time; Maximum temperature rise at the self-healing point; Dissipation factor of voltage variation; The intensity of voltage fluctuations; Fractal dimension of the fracture surface of the self-healing point.

[0013] Preferably, the classification model includes: The extracted classification features are used as input to the classification model; Based on the classification characteristics, a category distribution model is established. The distribution of each category is determined by the central feature value and the feature variation range. The category distribution model is dynamically updated based on historical data and newly collected data to optimize the classification accuracy of the classification model. The category with the highest output matching degree is taken as the classification result of the initial self-healing type of the metallized film capacitor.

[0014] Preferably, the metallized film capacitor initial self-healing type classification system includes the following modules: A boost circuit module is used to apply voltages with different boost rates to the metallized film capacitor under test; The data acquisition module is used to collect the self-healing voltage, power fluctuation data during the self-healing process, temperature rise at the self-healing point, and self-healing time range; the microscopic imaging module is used to obtain images of the fracture surface; Thermodynamic modeling module, used to build energy dissipation models and calculate maximum temperature rise; Dynamic modeling module, used to build stochastic process dynamics models and extract dissipation coefficients and fluctuation intensities; Fractal analysis module, used to calculate the fractal dimension of the fracture surface; The classification module is used to input the classification features into the classification model and output the initial self-healing type classification result of the metallized film capacitor.

[0015] The present invention provides a method for classifying the initial self-healing types of metallized film capacitors. This method has the following beneficial effects: 1. This invention achieves the technical effect of accurately classifying the initial self-healing type of metallized film capacitors by combining multidimensional feature extraction with Gaussian mixture model classification. Compared with existing classification solutions that rely on single features or manual judgment, this solves the problems of insufficient classification accuracy, difficulty in quantification, and poor repeatability.

[0016] 2. By combining thermodynamic modeling, kinetic modeling, and fractal geometry analysis, this invention quantifies the energy dissipation intensity, voltage decay trends and fluctuation characteristics, and the geometric complexity of the fracture surface during the self-healing process, achieving a comprehensive quantitative analysis of self-healing behavior. Compared to existing solutions that analyze only a single characteristic, this approach overcomes the lack of comprehensive analysis of multidimensional features.

[0017] 3. By employing high-frequency sampling and high-resolution imaging, this invention achieves high-precision and reliable experimental data collection. The data acquisition module records instantaneous voltage, current, and power fluctuations during the self-healing process, while the microscopic imaging module captures high-resolution images of the fracture surface. This approach overcomes the existing issues of inability to fully capture rapidly changing signals and fracture geometry, compared to existing solutions with lower sampling frequencies and susceptible data recording to interference.

[0018] 4. This invention achieves the technical effect of improving capacitor reliability assessment capabilities by accurately classifying and analyzing the initial self-healing types of metallized film capacitors and employing scientific modeling and classification methods. Compared to existing solutions that evaluate capacitor performance using only a single test method, this solves the problem of a single evaluation metric and difficulty in accurately predicting service life. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 is a flow chart of the method steps of the present invention; Figure 2 This is a system architecture diagram of the present invention. DETAILED DESCRIPTION

[0020] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the present specification. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0021] Please see the attached Figure 1 The present invention provides a method for classifying the initial self-healing types of metallized film capacitors. This method, based on multidimensional data feature analysis and the construction of a classification model, includes data acquisition and experimental design, energy dissipation model construction, temperature rise calculation, random process dynamics model establishment, fracture surface fractal geometry analysis, and classification model implementation. This method achieves accurate classification of the initial self-healing types of metallized film capacitors, thereby improving the reliability and safety of capacitor use. The method comprises the following steps: S1. Apply boost voltage and collect key data related to self-healing; S2. Construct an energy dissipation model and calculate the dissipation rate; S3. Determine the maximum temperature rise at the self-healing point based on the dissipation rate; S4. Establish a dynamic model to extract the dissipation coefficient and fluctuation intensity; S5. Calculate the fractal dimension and extract the fracture complexity characteristics; S6. Use the classification model to classify the self-healing types.

[0022] For step S1, multidimensional data on self-healing behavior is obtained through high-precision experiments and real-time monitoring, providing reliable data support for subsequent energy dissipation models, temperature rise analysis, random process dynamics modeling, and fracture fractal geometry analysis. Generally, the collected data should include key information such as the voltage at which self-healing occurs, the instantaneous power and current fluctuations during the self-healing process, the temperature rise at the self-healing point, and images of the fracture surface at the self-healing point. This data needs to be collected under precise experimental conditions using high-frequency sampling equipment and dedicated instruments, including the configuration of the experimental device, the setting of experimental parameters, and the collection and recording of data.

[0023] In this embodiment, under normal circumstances, the boost circuit module can provide a voltage output from 0V to 2000V and supports different boost rates, such as 0.5V / s, 5V / s, and 10V / s. As an option, the boost rate setting can be adjusted according to specific test requirements to simulate different actual working conditions. For example, low-speed boosting can be used to study self-healing behavior under long-term electric field effects, while high-speed boosting is suitable for testing self-healing characteristics under extreme working conditions.

[0024] In one possible implementation, the data acquisition module's sampling frequency is set to one million times per second to ensure that the instantaneous fluctuation characteristics of voltage and current during the self-healing process can be captured. Specifically, the collected data includes: the voltage value when self-healing occurs, the instantaneous current and power fluctuation signals during the self-healing process, and the power is calculated using the following formula: P(t)=V(t)·I(t) Where P(t) is the instantaneous power, V(t) is the instantaneous voltage, and I(t) is the instantaneous current; The temperature rise distribution of the self-healing point is collected by a thermal imager, and the temperature field data is recorded, and the maximum temperature rise value of the self-healing point is recorded; the geometric characteristic image of the fracture surface of the self-healing point is recorded by a microscopic imaging module for subsequent fractal geometry analysis.

[0025] Specifically, during the test, a metallized film capacitor was connected to a boost circuit module, and the voltage was gradually increased until self-healing occurred. Generally, self-healing can be detected by current fluctuations and power surges, while a thermal imager was used to capture temperature changes at the self-healing point. In one experiment, when the boost rate was set to 5 volts per second, a self-healing voltage of 1200 volts, a peak instantaneous power of 5 watts, a self-healing time ranging from 5 to 8 milliseconds, and a maximum temperature rise of 150 degrees Celsius were recorded.

[0026] As a possible implementation, the collection of temperature-rise data at the self-healing point requires combining the real-time recording capabilities of a thermal imager with time-series analysis to extract temperature trends. For example, in experiments, the temperature-rise data curve shows that the temperature at the self-healing point rises rapidly at the onset of healing and then gradually returns to its initial state after reaching its maximum temperature rise. By integrating the temperature-rise curve, the total heat released during the healing process can be calculated.

[0027] In an expanded implementation, to obtain more information about the fracture characteristics, further electron microscopy analysis can be performed on the self-healing points after the experiment to extract multi-level geometric features of the fracture surface. For example, scanning electron microscopy can reveal localized melting on the fracture surface, which is closely related to the arc discharge process and can provide supplementary data for subsequent fractal geometry analysis of the fracture.

[0028] It is important to note that during the experiment, the data acquisition timing should be synchronized with the changes in the boost signal to ensure that all key information is fully recorded. For example, when the boost rate is set to 10 volts per second, the data acquisition system needs to record the dynamic changes in voltage, current, power, and temperature rise data in real time during each boost phase.

[0029] In this embodiment, the collected data provides multi-dimensional feature inputs for subsequent steps. For example, the self-healing voltage and instantaneous power fluctuation signals are used to construct an energy dissipation model; temperature rise data are used to analyze local thermal effects during the self-healing process; and image data of the fracture surface are used to calculate the fractal dimension and quantify the geometric complexity of the fracture.

[0030] For step S2, by introducing the energy dissipation model and calculating the energy release rate per unit time, the intensity difference of the self-healing behavior is scientifically described to further quantify the energy release characteristics during the self-healing process of the metallized film capacitor.

[0031] Typically, when self-healing occurs, a localized arc discharge occurs at the self-healing point of a metallized film capacitor, releasing significant heat and electrical energy. This energy is dissipated into the surrounding environment within a short period of time. To effectively characterize this transient process, this paper employs an energy dissipation model based on power integration to calculate the average energy dissipation rate per unit time.

[0032] Specifically, the energy dissipation rate Φ is the ratio of the total energy released by the self-healing point within the self-healing time range to the length of the self-healing time. Its mathematical expression is: Where Φ represents the energy release rate per unit time during the self-healing process, t0 represents the time point when the self-healing process starts, P(t) represents the instantaneous power at time t, and t cIndicates the time point when the self-healing process ends, dt indicates the time interval [t0,t c ] is a very small time increment within .

[0033] In one possible implementation, in order to obtain accurate P(t) data, the sampling frequency is set to one million times per second (1MHz) to ensure that the dynamic changes of voltage and current during the self-healing process can be accurately captured in a short time.

[0034] In some embodiments, the total energy dissipation is calculated by considering experimental conditions with different boost rates. Specifically, different boost rates significantly affect the energy release characteristics of the self-healing phenomenon. The energy dissipation model is grouped by boost rate. For example: Under the condition of a voltage boost rate of 0.5 V / s, the self-healing process is slow and the energy release curve P(t) is relatively stable; Under the conditions of boost rates of 5V / s and 10V / s, the self-healing process is more intense, and the power curve P(t) shows obvious peaks.

[0035] By comparing the Φ results at different boost rates, the influence of boost rate on the self-healing characteristics of metallized film capacitors can be further revealed.

[0036] In some embodiments, to improve calculation accuracy and adaptability, a fitting method for experimental data is introduced. Generally, the change in P(t) during the self-healing process is not always smooth and may be affected by transient fluctuations. As an implementation method, a fitting method can be used to smooth the power curve.

[0037] Specifically, the collected P(t) data can be fitted using cubic spline fitting or Gaussian distribution fitting methods: cubic spline fitting is suitable for approximately smooth power curves and can better retain the main characteristics of P(t); Gaussian distribution fitting is more suitable for situations where the power curve contains obvious peaks, and the main characteristics of energy release in the self-healing process can be directly extracted through peak parameters.

[0038] In one possible implementation, the energy dissipation rate is used in subsequent feature extraction and classification steps. The calculated Φ, a key characteristic of self-healing behavior, directly reflects the intensity of energy release. In subsequent steps, Φ is input into a classification model as a classification feature to distinguish different types of self-healing behavior.

[0039] In step S3, the constructed energy dissipation model can be used to determine the energy dissipation rate per unit time during the self-healing process, a crucial physical parameter. By further integrating the thermal conductivity of the metallized film material, the temperature rise at the self-healing point, particularly the maximum temperature rise, is calculated, providing key data support for subsequent classification feature extraction. The calculation of temperature rise is a crucial step in this invention, aiming to accurately characterize the intensity of local thermal effects during the self-healing process through thermodynamic modeling.

[0040] Typically, the energy released during the self-healing process is concentrated in a localized area of the metallized film, causing a sharp temperature increase in that area. The released energy is then transferred to the surrounding material through thermal conduction, reaching thermal equilibrium within a certain timeframe. The magnitude of the temperature rise depends on the energy released per unit time, the characteristics of the heat conduction path, and the thermal conductivity of the material.

[0041] Specifically, this step first uses a thermal imager to monitor the temperature changes of the self-healing point in real time and collects the time series of temperature rise data, especially the moment when the temperature reaches the peak during the self-healing process.

[0042] As an option, based on the thermal imaging data, combined with the energy dissipation rate per unit time Φ, the theoretical temperature rise value of the self-healing point is calculated using thermodynamic formulas. The calculation formula for the theoretical temperature rise is: Where ΔT represents the temperature rise, Φ represents the energy dissipation rate per unit time, and k is the thermal conductivity of the metallized film material. This formula allows for theoretical correction of actual temperature rise data obtained from thermal imaging monitoring, thereby improving the accuracy of temperature rise calculations.

[0043] In one possible implementation, the thermal conductivity coefficient k can be directly obtained from a material parameter manual or through experimental calibration. For example, for common metallized polypropylene film materials, the thermal conductivity coefficient is typically in the range of 0.1 W / (m·K) to 0.3 W / (m·K). For multilayer composite film materials, a weighted average of the thermal conductivity coefficients of each layer is required.

[0044] In some embodiments, considering that the geometry and heat dissipation characteristics of metallized film capacitors may affect the temperature rise distribution, finite element analysis can be used for further verification. For example, by establishing a two-dimensional or three-dimensional heat conduction model to simulate the dynamic changes in the temperature field around the self-healing point, the actual temperature rise distribution can be determined to be consistent with the theoretical formula.

[0045] In general, the temperature rise of the self-healing point is not only related to the energy release rate and the thermal conductivity of the material, but also to the area of the self-healing point and the energy release time. In one possible implementation, the area of the self-healing point can be measured by a microscopic imaging module, and the energy release time is determined according to the time range [t0, t c For the case where the self-healing point area is small but the energy release rate is high, the temperature rise may have a local peak.

[0046] In some embodiments, comparative analysis can also be conducted using experimental data from different boost rates. At lower boost rates, the temperature rise at the self-healing point is generally lower because energy is released relatively slowly and evenly. However, at higher boost rates, the temperature rise at the self-healing point may increase significantly, with the peak temperature occurring earlier. This phenomenon can be intuitively demonstrated through the temperature curve.

[0047] As an option, the temperature rise trend can be further extracted by fitting the temperature rise time series curve, such as calculating the peak time of temperature rise, heating rate and cooling rate, etc. These data can be used for subsequent optimization of classification features.

[0048] Specifically, the fitting of the temperature rise time series can be done in the following form: Where T(t) represents the temperature at any time t, T base Indicates the reference temperature, ΔT max is the maximum temperature rise value, e is the base of the natural logarithm, λ is the temperature rise attenuation coefficient, t peak Indicates the time when the temperature rise reaches its peak. By fitting this curve, the temperature rise characteristics of the self-healing point can be more comprehensively characterized.

[0049] The temperature rise calculation method in this step is widely applicable to different types of metallized film capacitors, such as single-layer, double-layer, or composite-layer film capacitors. In some embodiments, the ranges of thermal conductivity and area parameters can be adjusted based on the different material properties of the capacitors to ensure the adaptability and accuracy of the temperature rise calculation.

[0050] For step S4, by analyzing the variation pattern of voltage over time, two key characteristic parameters, voltage dissipation coefficient and voltage fluctuation intensity, are extracted to provide support for the subsequent classification model.

[0051] Generally speaking, the voltage change during the self-healing process is a complex dynamic process, manifesting as a superposition of deterministic dissipative behavior and random fluctuations. Therefore, the present invention establishes a stochastic differential equation to accurately describe the change of the self-healing voltage over time, thereby quantifying its characteristic parameters.

[0052] In this embodiment, the specific construction method of the stochastic process dynamics model is as follows: During the self-healing process, the voltage change can be expressed by the following stochastic differential equation: dU(t)=-αU(t)dt+σdW t Where U(t) is the time variation of the self-healing voltage, dU(t) represents the variation of the self-healing voltage U(t) within a small time interval of time t, α represents the decay rate of the voltage over time during the self-healing process, σ represents the intensity of random fluctuations in the self-healing voltage variation, αU(t) represents the deterministic dissipation term of the voltage, and dW t represents the increment of Brownian motion in a random process at time t.

[0053] In one possible implementation, the dissipation coefficient α can be obtained by fitting the curve of voltage variation over time, and its calculation formula is: Among them, α represents the decay rate of voltage over time during the self-healing process, dU(t) represents the change value of the self-healing voltage U(t) within a small time interval of time t, Indicates the rate of change of voltage over time.

[0054] Alternatively, the fluctuation intensity σ can be obtained by calculating the standard deviation of the self-healing voltage. Specifically, based on the self-healing voltage data verified by the high-frequency sampling system, the standard deviation of the voltage within the self-healing time range is statistically calculated to obtain the value of σ.

[0055] In some embodiments, data collection and processing can be performed as follows: Typically, a high-frequency sampling system is used to capture the self-healing voltage change data in real time at a sampling frequency of 1 MHz, ensuring a sufficiently high temporal resolution. The collected data is filtered and used to fit the voltage change curve and calculate the voltage fluctuation characteristics.

[0056] Specifically, based on the collected voltage change curve, the numerical fitting method is used to calculate the voltage decay rate over time, thereby extracting the dissipation coefficient α.

[0057] In one possible implementation, the data fitting range can be selected from the time period from the start of self-healing to the return to stability to ensure the accuracy of the fitting. As another implementation, when calculating the fluctuation intensity σ, all voltage data within the self-healing time range are selected and calculated using the following formula: Where n represents the total number of voltage data points collected within the selected time range; U i represents the i-th voltage value collected within the time range; is the mean of the sampled data, It represents the square of the deviation between the ith voltage data point and the voltage mean.

[0058] In specific applications, the random process dynamics model has the following characteristics: As a possible implementation method, this model can simultaneously capture the overall trend of voltage decay and the random characteristics of voltage fluctuations, which provides key parameter support for different types of self-healing behaviors.

[0059] In another implementation, the magnitude of the dissipation coefficient α can be used to distinguish between fast-decay and slow-decay types of self-healing behavior. Fast decay typically corresponds to high-energy self-healing behavior, while slow decay may correspond to low-energy self-healing behavior. Furthermore, the magnitude of the fluctuation intensity σ can reflect the magnitude of random fluctuations in the self-healing process. Large fluctuation intensities may be associated with high local stresses or complex arcing behavior within the material, while smaller fluctuation intensities may indicate more stable self-healing behavior.

[0060] In step S5, analysis of the fractal dimension of the fracture surface can further distinguish different types of self-healing behavior, providing an independent and important input parameter for the classification model. In this embodiment, the fractal dimension calculation based on the fracture surface image primarily uses the box counting method in fractal geometry. Specifically, the fractal dimension can quantify the complexity of the fracture surface and characterize the differences in the geometric characteristics of self-healing behavior, further supporting the accurate construction of the classification model.

[0061] Generally, a microscopic imaging module can capture clear images of the fracture surface. The magnification of the microscopic imaging module should meet the requirements for capturing fracture surface details, typically between 100x and 1000x. The resolution of the microscopic imaging image should reach 0.1 micron per pixel to accurately distinguish the fine structure of the surface fracture.

[0062] As an option, the collected images can be preprocessed to enhance their analytical effectiveness, including: grayscale processing: converting color images into grayscale images to highlight the surface geometric structure; denoising: using Gaussian filtering or median filtering to remove image noise; edge extraction: using the Canny operator to extract the edge structure of the fracture surface to improve the accuracy of fractal dimension calculation.

[0063] In one possible implementation, after image processing is completed, a binary image can be generated to clearly mark the main areas of the fracture surface for subsequent fractal dimension calculation.

[0064] In this embodiment, the fractal dimension of the fracture surface is calculated by the box counting method. Specifically, the fractal dimension D f The calculation formula is: Among them, D f It represents the geometric complexity of the fracture surface image, N(r) is the number of covering units with a scale of r on the fracture surface, r represents the unit scale used to cover the fracture surface, log(N(r)) represents the logarithm of the number of units N(r) required to cover the fracture surface, and log(1 / r) represents the logarithm of the reciprocal of the measurement scale r.

[0065] In general, in order to improve the accuracy of the calculation, we can select multiple groups of different covering unit side lengths r, such as r = 2, 4, 8, 16, 32, etc., calculate the number of covering units N(r) corresponding to each group, and use the least squares method to perform a linear fit on the relationship curve between log(N(r)) and log(1 / r) to obtain the fractal dimension D. f .

[0066] In some embodiments, in order to ensure that the calculation of the fractal dimension of the fracture surface can reflect the actual geometric characteristics, the number of covering units can be repeatedly measured and the average value can be selected as the final fractal dimension input data.

[0067] In step S6, based on the acquired multidimensional features, including energy dissipation rate, maximum temperature rise, dissipation coefficient of voltage variation, voltage fluctuation intensity, and fractal dimension of the fracture surface, accurate classification of the initial self-healing type of the metallized film capacitor is achieved through processing and analysis of these characteristic data. The construction and implementation of the classification model is a key technical solution of this invention. Its core lies in the use of mathematical modeling methods to comprehensively analyze the multidimensional features to obtain accurate self-healing type classification results.

[0068] The implementation of the classification model includes the following: As an implementation method, the classification model input features include feature data of the following five dimensions: Energy dissipation rate per unit time; Maximum temperature rise at the self-healing point; Dissipation factor of voltage variation; The intensity of voltage fluctuations; Fractal dimension of the fracture surface of the self-healing point.

[0069] Specifically, in one possible implementation, the classification model uses a Gaussian mixture model as the classification algorithm. GMM can effectively classify multidimensional features by assuming that data features conform to a mixture model of multiple Gaussian distributions. The classification model construction process is as follows: Construct a multidimensional feature vector based on the extracted feature data. In general, the multidimensional feature vector is expressed as: x=[Φ,ΔT max ,α,σ,Df ] Among them, Φ represents the energy dissipation rate per unit time, which is expressed by the formula Calculated, P(t) is the instantaneous power, [t0,t c ] is the self-healing time range, dt represents the small time interval; ΔT max Represents the maximum temperature rise of the self-healing point; α is the dissipation coefficient of voltage change, which is calculated as follows: α represents the decay rate of voltage over time during the self-healing process, dU(t) represents the change in the self-healing voltage U(t) within a small time interval of time t, Indicates the rate of change of voltage over time; D f The fractal dimension of the fracture surface is expressed by the formula It is calculated that N(r) is the number of covering units with a scale of r on the fracture surface, r represents the unit scale used to cover the fracture surface, log(N(r)) represents the logarithm of the number of units N(r) required to cover the fracture surface, and log(1 / r) represents the logarithm of the reciprocal of the measurement scale r.

[0070] In the construction of the classification model, the distribution of each category is determined by the central eigenvalue and the range of feature variation. Specifically, the classification model assumes that the data characteristics of each category conform to the Gaussian distribution, and its probability density function is expressed as: Among them, x represents the feature vector of the sample to be classified; θ k represents the Gaussian distribution parameter of category k, μ k Indicates the center position of the feature data of category k in each dimension; ∑ k Represents the correlation and distribution range of the feature data of category k between various dimensions; represents the covariance matrix ∑ k The inverse matrix, (x-μ k ) represents the mean vector μ of sample x and category k k The gap between them, d represents the dimension of the input sample feature vector x, (2π) d / 2 It is the normalization coefficient in the Gaussian distribution probability density formula, and exp is used to calculate the core exponential part of the probability density.

[0071] As an option, in the actual classification process, the classification model determines the category to which the sample belongs by maximizing the posterior probability. The posterior probability calculation formula is: Among them, x represents the feature vector of the sample to be classified, θ k represents the Gaussian distribution parameter of category k, P(C k|x) represents the posterior probability that the sample belongs to the category, P(x|θ k ) represents the characteristic distribution of the category, Represents the weighted sum of likelihood probability and prior probability under all categories, P(C k ) represents the prior probability of the category; K is the total number of categories, C j Represents the j-th category in the classification model.

[0072] In some embodiments, in order to improve classification accuracy and adaptability, the classification model introduces a dynamic update mechanism. Specifically, the classification model updates the parameters of the category distribution model based on historical data and newly collected data, including the mean vector μ k and the covariance matrix Σ k The dynamic update process is as follows: The newly collected data features are added as incremental data to the existing classification model; the parameters of each category are re-estimated based on the new data, and the parameters are iteratively updated through the expectation maximization (EM) algorithm.

[0073] As an implementation method, dynamic updating of the classification model can adapt to different batches of capacitor samples and improve the robustness of the classification results.

[0074] In the output of the classification model, in general, the classification model outputs the category with the highest degree of matching with the input features as the final self-healing type classification result. For example, when the input feature vector [x1, x2, ..., x d The posterior probability P(C k |x) in category C k When the maximum value is reached, the category C k As the classification result of the sample.

[0075] The following description of a classification system for initial self-healing types of metallized film capacitors and the above description of a distributed management method for a cloud container cluster may refer to each other.

[0076] Please see the attached Figure 2 The present invention also provides a classification system for the initial self-healing types of metallized film capacitors. This system combines high-frequency sampling technology, thermodynamic modeling, stochastic process dynamics modeling, and fractal geometry analysis, among other technical approaches. Through a modular system architecture, relying on a boost circuit module, a data acquisition module, a microscopic imaging module, a thermodynamic modeling module, a dynamics modeling module, a fractal analysis module, and a classification module, the system can collect, process, and classify multidimensional data from the self-healing process of metallized film capacitors. This system can scientifically classify the initial self-healing types of metallized film capacitors, providing important technical support for capacitor reliability assessment and life prediction.

[0077] The boost circuit module applies voltages at varying rates to the metallized film capacitors under test, providing an adjustable voltage output from 0 to 2000 volts and supporting multiple boost rates (e.g., 0.5 volts per second, 5 volts per second, and 10 volts per second). This module precisely controls the voltage ramp rate, ensuring that voltage changes during testing meet predetermined experimental parameters.

[0078] In this implementation, the boost circuit module includes a programmable power controller and a precision voltage regulator. It dynamically adjusts the output voltage through closed-loop feedback technology, ensuring accurate and consistent boost rates. The controller automatically adjusts the voltage ramp rate according to preset parameters and monitors changes in the electrical signal generated by the capacitor during the boost process to trigger subsequent data acquisition. The boost circuit module provides a stable and controllable boost environment, enabling repeatable and reliable triggering and analysis of self-healing behavior.

[0079] The data acquisition module is used to collect multi-dimensional dynamic data in the self-healing process in real time, including key data such as instantaneous voltage, self-healing voltage, self-healing current, self-healing time range, and self-healing power fluctuation.

[0080] In this implementation, the data acquisition module uses high-speed sampling technology to achieve high-precision signal capture through voltage and current detection units. This module generates a raw data stream containing voltage, current, and time series data at a sampling rate of millions per second. A clock synchronization mechanism within the detection units ensures the timing consistency of self-healing events. The collected data is transmitted to the modeling module via a digital interface for subsequent processing.

[0081] The data acquisition module can record rapidly changing self-healing dynamic signals with high precision and without delay, providing a complete data basis for feature extraction and analysis.

[0082] The microscopic imaging module is used to collect fracture surface images of the self-healing point and record the fracture characteristics generated during the self-healing process.

[0083] In this implementation, the microscopic imaging module combines a high-resolution microscope with an image acquisition system, using autofocus and image enhancement techniques to produce high-precision images of the fracture surface. The microscope's resolution is sufficient to capture micron-level fracture details and enhance key geometric features in the image, making subsequent analysis more reliable.

[0084] The microscopy imaging module provides clear and complete fracture surface data for fractal geometry analysis, helping to accurately quantify the complexity of geometric features.

[0085] The thermodynamic modeling module is used to analyze the dissipation and conversion of energy during the self-healing process. The total energy released during the self-healing process and the dissipation rate per unit time are determined through energy analysis.

[0086] In this implementation, the thermodynamic modeling module detects the voltage and current signals during self-healing and calculates the energy released per unit time. This module uses a combination of integration techniques and dynamic data smoothing algorithms to extract the total energy dissipated during self-healing behavior. Combined with material thermal conductivity parameters, it estimates the local temperature rise caused by this energy release. Furthermore, based on the changing patterns of multiple data sets, the module generates a dissipation rate curve to reflect the intensity characteristics of the self-healing behavior.

[0087] The thermodynamic modeling module effectively quantifies the energy release intensity during the self-healing process and provides core parameters for subsequent classification feature construction.

[0088] The dynamic modeling module is used to establish a random process dynamic model and analyze the trend and fluctuation characteristics of voltage changes during the self-healing process.

[0089] In this implementation, the kinetic modeling module uses a fitting algorithm to extract the steady decay rate of voltage changes based on the time series of voltage changes. It also uses statistical analysis of the voltage signal to assess its fluctuation intensity. This module employs a customized multi-order fitting method to extract the main trends of voltage decay from large-scale data. Combined with the distribution characteristics of voltage fluctuation amplitudes, this module generates dynamic characteristic indicators to describe self-healing behavior.

[0090] The kinetic modeling module reveals the attenuation characteristics and random fluctuation features of voltage changes, providing the necessary kinetic information for the construction of the classification model.

[0091] The fractal analysis module is used to calculate the geometric complexity of the fracture surface and provide analysis results based on geometric features for self-healing type classification.

[0092] In this implementation, the fractal analysis module, combined with fracture surface images captured by the microscopic imaging module, uses digital image segmentation techniques to segment the surface area. By counting geometric overlaps at different scales, the distribution characteristics of the surface geometry are extracted. The module's pre-defined fractal geometry algorithm quantifies surface complexity, deriving numerical indices describing geometric properties. These indices are then analyzed for distribution and subsequent classification.

[0093] The fractal analysis module quantifies the geometric characteristics of the fracture surface into numerical parameters, providing a reliable basis for the classification of self-healing types.

[0094] The classification module is used to perform comprehensive analysis on the extracted feature data to classify the initial self-healing type of metallized film capacitors.

[0095] In this implementation, the classification module uses the input feature data as a basis to construct a classifier using a Gaussian mixture model (GMM). This module performs matching analysis on the input multidimensional feature data, calculates the posterior probability based on the similarity between the feature data and each class distribution, and outputs the optimal classification result. The classification module supports a dynamic update mechanism that continuously introduces new data to optimize the parameters of the classification model, including the class center and the range of the class distribution, thereby improving the adaptability and accuracy of the classification.

[0096] The classification module integrates multi-dimensional feature data to accurately determine the initial self-healing type, providing important technical support for the reliability assessment of capacitors.

[0097] The system of this embodiment can be used to execute the above method embodiments, and its principles and technical effects are similar, so they will not be repeated here.

[0098] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A method for classifying the initial self-healing type of metallized film capacitors, characterized in that: The following steps are involved: Apply voltages at different ramp rates to the metallized film capacitor under test and collect key data during self-healing, including self-healing voltage, power fluctuation data during the self-healing process, temperature rise at the self-healing point, self-healing time range, and images of the fracture surface at the self-healing point. Based on the self-healing voltage and the power fluctuation data during the self-healing process, an energy dissipation model is constructed to calculate the energy dissipation rate per unit time; Determine the maximum temperature rise of the self-healing point based on the energy dissipation rate and the thermal conductivity of the material; Establishing a random process dynamics model for the change of the self-healing voltage over time, and extracting the dissipation coefficient and fluctuation intensity of the self-healing voltage change; Based on the fracture surface image of the self-healing point, the fractal dimension of the fracture surface is calculated and the fracture complexity characteristics of the self-healing point are extracted; The initial self-healing type of metallized film capacitors is classified using a classification model based on classification features.

2. The method for classifying the initial self-healing type of metallized film capacitors according to claim 1, characterized in that: The voltage range of the different voltage increasing rates applied to the metallized film capacitor to be tested is 0.5 volts per second to 10 volts per second, the applied voltage range is 0 to 2,000 volts, and the collected data includes the following: The voltage value when self-healing occurs; Instantaneous voltage and current data during the self-healing process; Self-healing power fluctuation curve; Time series changes of temperature rise at the self-healing point; The start and end time of self-healing.

3. Among them, the data acquisition frequency is one million times per second, which can meet the needs of high-precision dynamic change capture.

4. The method for classifying the initial self-healing type of metallized film capacitors according to claim 1, characterized in that: The energy dissipation rate per unit time is calculated by the following formula: Where, φ represents the energy release rate per unit time during the self-healing process, t0 represents the time point when the self-healing process starts, P(t) represents the instantaneous power at time t, and t c Indicates the time point when the self-healing process ends, dt indicates the time interval [t0,t c ] is a very small time increment within .

5. The method for classifying the initial self-healing type of metallized film capacitors according to claim 1, characterized in that: The maximum temperature rise of the self-healing point is determined by the following process: Obtain energy release data during the self-healing process, including energy dissipation per unit time; According to the thermal conductivity of the metallized film capacitor material, the energy released per unit time is converted into temperature rise; The temperature change of the self-healing point is collected in real time, and its maximum temperature rise value is recorded. The temperature rise value reflects the heat dissipation characteristics of the metallized film capacitor material during the self-healing process; Temperature rise data combined with material characteristic parameters are used as important feature quantities for subsequent classification models.

6. The method for classifying the initial self-healing type of metallized film capacitors according to claim 1, characterized in that: The stochastic process dynamics model is described by the following stochastic differential equation: dU(t)=-αU(t)dt+σdW t Where U(t) is the time variation of the self-healing voltage, dU(t) represents the variation of the self-healing voltage U(t) within a small time interval of time t, α represents the decay rate of the voltage over time during the self-healing process, σ represents the intensity of random fluctuations in the self-healing voltage variation, αU(t) represents the deterministic dissipation term of the voltage, and dW t represents the increment of Brownian motion in a random process at time t.

7. The method for classifying the initial self-healing type of metallized film capacitors according to claim 1, characterized in that: The stochastic process dynamics model extracts the dissipation coefficient and fluctuation intensity through the following steps: The dissipation coefficient is calculated by the following formula: Among them, α represents the decay rate of voltage over time during the self-healing process, dU(t) represents the change value of the self-healing voltage U(t) within a small time interval of time t, Indicates the rate of change of voltage over time; The fluctuation intensity is extracted through statistical analysis of the self-healing voltage, specifically the standard deviation of the self-healing voltage; The data source includes the dynamic capture of the self-healing voltage by a high-frequency sampling system with a sampling frequency of one million times per second.

8. The method for classifying the initial self-healing type of metallized film capacitors according to claim 1, characterized in that: The fractal dimension of the fracture surface image is calculated by the following formula: Among them, D f It represents the geometric complexity of the fracture surface image, N(r) is the number of covering units with a scale of r on the fracture surface, r represents the unit scale used to cover the fracture surface, log(N(r)) represents the logarithm of the number of units N(r) required to cover the fracture surface, and log(1 / r) represents the logarithm of the reciprocal of the measurement scale r.

9. The method for classifying the initial self-healing type of metallized film capacitors according to claim 1, characterized in that: The classification features include the following five main features: Energy dissipation rate per unit time; Maximum temperature rise at the self-healing point; Dissipation factor of voltage variation; The intensity of voltage fluctuations; Fractal dimension of the fracture surface of the self-healing point.

10. The method for classifying the initial self-healing type of metallized film capacitors according to claim 1, characterized in that: The classification model includes: The extracted classification features are used as input to the classification model; According to the classification characteristics, a category distribution model is established. The distribution of each category is determined by the central feature value and the feature variation range; Dynamically update the category distribution model based on historical data and newly collected data to optimize the classification accuracy of the classification model; The category with the highest output matching degree is taken as the classification result of the initial self-healing type of the metallized film capacitor.

11. A classification system for initial self-healing types of metallized film capacitors, applied to a classification method for initial self-healing types of metallized film capacitors as claimed in any one of claims 1 to 9, characterized in that: The initial self-healing type classification system for metallized film capacitors includes the following modules: A boost circuit module is used to apply voltages with different boost rates to the metallized film capacitor under test; Data acquisition module, used to collect self-healing voltage, power fluctuation data during self-healing process, temperature rise at self-healing point and self-healing time range; A microscopic imaging module for acquiring images of the fracture surface; Thermodynamic modeling module, used to build energy dissipation models and calculate maximum temperature rise; Dynamic modeling module, used to build stochastic process dynamics models and extract dissipation coefficients and fluctuation intensities; Fractal analysis module, used to calculate the fractal dimension of the fracture surface; The classification module is used to input the classification features into the classification model and output the initial self-healing type classification result of the metallized film capacitor.

Images