Method for monitoring service life of metallized film capacitor

Abstract

The invention relates to the technical field of power electronics, and discloses a method for monitoring the service life of a metallized film capacitor, which comprises the following steps: collecting multi-physical field signals of the metallized film capacitor in real time through a sensor, preprocessing the collected multi-physical field signals, establishing a coupling model based on multiple physical fields, and calculating the service life of the metallized film capacitor. The method comprises the following steps: obtaining multi-physical field distribution of a metallized film capacitor, carrying out feature extraction on the multi-physical field distribution through wavelet transform, calculating multi-scale energy distribution characteristics of signals, carrying out dynamic reasoning on the health state of the capacitor based on the multi-scale energy distribution characteristics in combination with a dynamic Bayes network, and evaluating the change rule of the health state of the capacitor. And based on the health state change rule, establishing a life loss model, and generating a health assessment report. Multi-physical field signals are collected in real time, physical field distribution characteristics in the capacitor are analyzed, the health state of the capacitor is inferred in combination with a dynamic Bayes network, the remaining life is calculated, and a health assessment report is generated.

Description

Technical Field

[0001] The present invention relates to the field of power electronics technology, and particularly to a method for monitoring the life of a metallized film capacitor. Background Art

[0002] As a key component in high-voltage direct current transmission systems and pulsed power systems, metallized film capacitors are widely used in extreme environments due to their high capacitance, excellent insulation performance, and self-healing characteristics. However, under complex working conditions of high electric fields, large temperature rises, and frequent charge and discharge, the internal dielectric of the capacitor gradually ages, partial discharges occur frequently, and at the same time, the accumulation of thermal effects and mechanical stresses leads to a gradual decline in the performance of the capacitor. This performance degradation process caused by the coupling of multiple factors poses a great challenge to the monitoring of the operating state and life prediction of the capacitor.

[0003] In the prior art, most methods evaluate the health state of capacitors through single physical field signals. For example, the method of off-line measuring capacitance values or temperature rises is used to indirectly reflect the aging degree of capacitors, or the characteristic signals of partial discharges are monitored to infer the deterioration of the dielectric. This monitoring method based on a single physical field lacks comprehensiveness because the aging of capacitors is caused by the interaction of multiple physical fields such as electric fields, thermal fields, and mechanical stresses, and a single signal cannot accurately reflect the complex physical behavior inside. In addition, since these methods usually use off-line detection or periodic monitoring means, the real-time evaluation of the operating state of capacitors cannot be achieved.

[0004] Traditional signal analysis methods also have deficiencies in feature extraction. For example, the Fourier transform can only obtain the spectral information of signals and cannot describe the local characteristics of signals in the time and frequency domains, which is far from enough for the analysis of non-stationary signals during the operation of metallized film capacitors. In addition, time-domain statistical analysis methods often only focus on simple features such as the amplitude or mean value of signals and cannot deeply capture the multi-scale complexity of signals changing with time. The limitations of these technical means lead to the loss of important feature information during the health state evaluation process, making the judgment of the health state of capacitors lack accuracy and pertinence. Summary of the Invention

[0005] Aiming at the deficiencies of the prior art, the present invention provides a method for monitoring the life of a metallized film capacitor, which solves the problem that in the prior art, the health state of a metallized film capacitor cannot be accurately and real-time monitored based on comprehensive multi-physical field analysis and its remaining life cannot be predicted.

[0006] To achieve the above object, the present invention is realized through the following technical solutions: A method for monitoring the life of a metallized film capacitor includes the following steps: Collect multi - physical field signals of metallized film capacitors in real - time through sensors, and pre - process the collected multi - physical field signals; establish a coupling model based on the multi - physical fields to obtain the multi - physical field distribution of metallized film capacitors; Extract features from the multi - physical field distribution through wavelet transform, and calculate the multi - scale energy distribution characteristics of the signals; Based on the multi - scale energy distribution characteristics, combine with the dynamic Bayes network to perform dynamic inference on the health state of the capacitor, and evaluate the changing law of its health state; Based on the changing law of the health state, establish a life loss model, calculate the remaining life of the metallized film capacitor, and generate a health assessment report.

[0007] Preferably, collecting multi - physical field signals includes the following steps: The electric field signal is measured by an electric field probe, and the electric field strength is calculated through an electric field distribution model; The thermal field signal is measured by a thermocouple or an infrared thermal imager, and the temperature distribution is calculated through a heat conduction model; The mechanical stress signal is measured by a piezoelectric sensor, and the stress distribution is calculated through a coupling model of thermal stress and electric field force; The partial discharge signal is obtained by a high - speed data acquisition device, and the sampling rate is not less than 1 GHz.

[0008] Preferably, the coupling model is established based on the following relationships: The electric field distribution is described by Poisson's equation; The thermal field distribution is described by the heat conduction equation, considering the heat source caused by partial discharge; The force field distribution is described by the combined action of thermal stress and electric field force.

[0009] Preferably, the pre - processing of multi - physical field signals includes: Adopt low - pass filtering to remove high - frequency noise; Normalize the electric field signal, thermal field signal and mechanical stress signal respectively; Synchronize the signals through a time alignment method to ensure the time consistency of multi - physical field signals.

[0010] Preferably, the wavelet transform adopts continuous wavelet transform, and the signal is decomposed into multiple scales through wavelet basis functions to extract the energy distribution characteristics on different time scales.

[0011] Preferably, the multi - scale energy distribution characteristics are used to calculate the energy entropy of multi - physical field signals, and the calculation method of energy entropy is: Perform wavelet decomposition on the collected multi - physical field signals, perform multi - scale decomposition on the signal x(t) using wavelet basis functions, and the expression of wavelet transform is: Among them, W(a, b) is the wavelet coefficient, a is the scale factor, and b is the time translation factor, which is the wavelet basis function; Calculate the corresponding energy value for each scale wavelet coefficient obtained by wavelet decomposition. The calculation formula for the energy value is: E i =|W(a i ,b)| 2 Among them, E i represents the wavelet energy value of the i-th scale; Normalize the wavelet energy values of each scale to obtain the wavelet energy ratio of each scale. The normalization formula is: Among them, P i is the wavelet energy ratio of the i-th scale, E i is the energy value of the signal at the i-th scale, N is the total number of decomposition scales, is the total wavelet energy of all scales and is used for normalization processing; Calculate the wavelet energy entropy. The expression of the wavelet energy entropy is: Among them, H represents the wavelet energy entropy of the multi-physical field signal, reflecting the complexity and chaos degree of the signal; Use the wavelet energy entropy as the signal feature and input it into the life prediction model to characterize the dynamic change characteristics of the multi-physical field signal.

[0012] Preferably, the construction of the dynamic Bayes network includes the following: Define the system health state and observation variables. The system health state follows a state transition probability distribution over time, and the state transition probability is related to the discharge intensity and aging rate; The relationship between the observation variables and the health state is described by the conditional probability distribution; Based on the eigenvalue of the multi-physical field signal observed in real time, use the dynamic inference method to estimate the system health state.

[0013] Preferably, the life loss model is established based on the cumulative change amount of the energy entropy of the multi-physical field signal. The larger the cumulative change amount, the faster the life decay rate.

[0014] Preferably, the life loss model is described by the following relationship: The remaining life is the reciprocal function of the initial life and the cumulative change amount of the energy entropy; When the cumulative change amount of the energy entropy exceeds the preset threshold, the predicted value of the remaining life decreases significantly.

[0015] Preferably, the health assessment report includes the current health status of the metallized film capacitor, the prediction result of the remaining life, and a warning prompt on whether maintenance is required. When the remaining life is lower than the set threshold, the system generates an alarm signal.

[0016] The present invention provides a method for monitoring the life of a metallized film capacitor. It has the following beneficial effects: 1. The present invention adopts a technical solution of multi-physical field signal acquisition and coupled modeling. By collecting electric field, thermal field, mechanical stress field, and partial discharge signals in real time, and constructing a coupling model between the electric field, thermal field, and force field, it achieves the technical effect of comprehensively quantifying the complex physical phenomena inside the metallized film capacitor. Compared with the prior art that only relies on a single physical field signal for health status assessment, it solves the problem of insufficient assessment accuracy caused by ignoring the multi-field interaction.

[0017] 2. The present invention adopts a technical solution of feature extraction based on wavelet transform. By multi-scale decomposition, it extracts the dynamic energy distribution characteristics of multi-physical field signals, and uses wavelet energy entropy to characterize the complexity of the signals, achieving the technical effect of accurately capturing the variation law of the signals at different time scales. Compared with the prior art that only uses time-domain or frequency-domain analysis methods to extract single features, it solves the problems of insufficient feature extraction and poor ability to capture the dynamic characteristics of the signals.

[0018] 3. The present invention adopts a technical solution of dynamic Bayesian network for dynamic inference of health status. Combining the health status transition model and the observation model, it realizes the real-time inference and dynamic update of the capacitor's health status, achieving the technical effect of accurately evaluating the evolution law of the health status over time. Compared with the prior art that relies on a static assessment model for health diagnosis, it solves the problems of being unable to track the change of the health status in real time and the lag of the assessment result.

[0019] 4. The present invention adopts a life loss model based on the cumulative change of coupling entropy. Combining the dynamic inference result of the health status, it calculates the remaining life of the metallized film capacitor and generates a health assessment report, achieving the technical effect of accurately predicting the capacitor's life and providing maintenance suggestions. Compared with the prior art that relies on empirical formulas or fixed life cycles for assessment, it solves the problem of being unable to perform personalized life prediction according to the specific operating status of the equipment. Description of the Drawings

[0020] Figure 1 It is a step diagram of the method for monitoring the life of the metallized film capacitor of the present invention Detailed Embodiments

[0021] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0022] Please refer to the attached Figure 1 , the embodiments of the present invention provide a method for monitoring the life of a metallized film capacitor, including the following steps: S1. Real-time collect multi-physical field signals of the metallized film capacitor through a sensor, and preprocess the collected multi-physical field signals; S2. Establish a coupling model based on the multi-physical fields to obtain the multi-physical field distribution of the metallized film capacitor; S3. Extract features from the multi-physical field distribution through wavelet transform, and calculate the multi-scale energy distribution characteristics of the signal; S4. Based on the multi-scale energy distribution characteristics, combine with a dynamic Bayes network to dynamically infer the health state of the capacitor, and evaluate the change law of its health state; S5. Based on the change law of the health state, establish a life loss model, calculate the remaining life of the metallized film capacitor, and generate a health assessment report.

[0023] In this embodiment, for step S1, first, the present invention obtains basic data reflecting the internal state and performance changes of the capacitor by real-time collecting and preprocessing the multi-physical field signals of the metallized film capacitor in the operating state. These signals can effectively describe the aging process, partial discharge behavior, thermal effect and stress distribution of the capacitor, and provide a reliable input data source for subsequent multi-physical field coupling analysis and life prediction.

[0024] Generally, the multi-physical field signals include electric field signals, thermal field signals, mechanical stress signals and partial discharge signals. For each type of signal collection, appropriate collection methods and devices need to be selected in combination with the specific operating environment and working state of the capacitor. To ensure the accuracy of the signals, the collected original signals need to be preprocessed, including but not limited to denoising, normalization and time synchronization processing, and finally provide a unified high-quality data stream as the basic input for subsequent analysis. Use various types of high-precision sensors arranged inside or outside the metallized film capacitor to real-time collect the multi-physical field signals of the capacitor during operation. The specific content is as follows: The electric field signal is obtained through a high-precision electric field probe. The electric field probe can be arranged near the internal dielectric of the capacitor to detect the electric field strength distribution of the capacitor under the operating voltage. The electric field signal can be calculated by the following formula: Among them, E is the electric field strength with the unit of volts per meter, and φ is the electric potential with the unit of volts. The electric field strength distribution satisfies the Poisson equation: Among them, ∈ is the permittivity with the unit of farads per meter, and ρ is the charge density with the unit of coulombs per cubic meter.

[0025] As an option, in actual acquisition, the arrangement position and sampling frequency of the electric field probe need to be optimized according to the specific size and working voltage of the capacitor to ensure capturing high-precision electric field signals.

[0026] The thermal field signal is obtained through a thermocouple or an infrared thermal imager and is used to monitor the temperature distribution during the operation of the capacitor. The thermal field signal can reflect the internal thermal loss and heat dissipation capacity of the capacitor, and its distribution can be described by the heat conduction equation as follows: Among them, T is the temperature with the unit of Kelvin, ρ is the material density with the unit of kilograms per cubic meter, C p is the specific heat capacity of the material with the unit of joules per kilogram per Kelvin, k is the thermal conductivity of the material with the unit of watts per meter per Kelvin, Q is the heat source term with the unit of watts per cubic meter, which is determined by the thermal loss caused by partial discharge.

[0027] In a possible implementation, the thermocouple is installed on the surface of the capacitor and at key heat-generating points to obtain real-time temperature distribution data in different regions. The infrared thermal imager can provide an overall thermal field distribution map of the capacitor, and specific temperature change data can be extracted through image processing algorithms.

[0028] The mechanical stress signal is obtained through a piezoelectric sensor. The mechanical stress signal can reflect the strain behavior of the capacitor dielectric and film materials caused by thermal expansion and electric field force. The stress distribution satisfies the following mechanical equilibrium equation: Among them, σ is the stress tensor with the unit of pascals, and f is the body force density with the unit of newtons per cubic meter.

[0029] In some embodiments, the piezoelectric sensor is arranged in key areas of the capacitor film structure, such as the edge-fixed area or the force-concentrated area, to capture dynamic stress distribution data.

[0030] The partial discharge signal is collected through a high-speed data acquisition system. Generally, the partial discharge signal is a high-frequency pulse signal, and the sampling rate needs to be not less than 1 GHz to ensure capturing the main characteristics of the pulse signal.

[0031] Specifically, the key parameters of partial discharge signals include amplitude, duration, repetition frequency, etc. These parameters can be further subjected to feature extraction after being collected by a high-speed ADC. For example, in the time domain, the discharge amplitude can be analyzed through peak detection, while in the frequency domain, the spectral characteristics of the discharge signal can be extracted through Fourier transform.

[0032] In this embodiment, after signal acquisition is completed, preprocessing of multi-physical field signals is required to improve the quality and consistency of the data. The specific content is as follows: First, noise reduction processing is performed on the collected signals. Generally, a low-pass filter is used to filter out high-frequency noise in the signals. As an option, according to the spectral characteristics of partial discharge signals, the cut-off frequency of the low-pass filter can be set within a range higher than the main signal components, such as a frequency higher than 10 MHz.

[0033] Secondly, normalization processing is performed on the signals. The purpose of normalization processing is to convert physical signals of different types and different dimensions into unified dimensionless values. The normalization formula is as follows: where x norm is the normalized signal value, x is the original signal value, x min and x max are the minimum and maximum values of the signal respectively.

[0034] Finally, time synchronization processing is performed on the signals. Generally, by setting a unified time stamp reference for each type of signal, time alignment is achieved. In one possible implementation, the acquisition system can provide a standard time series through the master clock, align the data collected by different sensors according to the time stamp, and ensure that the multi-physical field signals have a consistent time axis.

[0035] Through the above signal acquisition and preprocessing steps, it can be ensured that the collected multi-physical field signals have high precision, high consistency, and usability, providing reliable data input for subsequent multi-physical field coupling modeling and feature extraction.

[0036] In this embodiment, for step S2, in order to accurately describe the complex physical phenomena of the metallized film capacitor in the operating state, after collecting and preprocessing the multi-physical field signals, the present invention establishes a coupling model based on the multi-physical field signals to analyze the distribution characteristics of the electric field, thermal field, and mechanical stress field inside the capacitor and the interaction relationship between them.

[0037] In general, multi-physics coupling modeling needs to simultaneously consider the mutual coupling relationships among the electric field distribution, heat conduction behavior, and mechanical stress distribution. When specifically implemented, the dynamic behaviors of each physical field are described by constructing an electric field model, a thermal field model, and a force field model, and the solutions are obtained by combining the actual boundary conditions. Finally, the multi-physics distribution inside the metallized film capacitor is obtained. The specific implementation details are as follows: In this embodiment, the electric field model uses the Poisson equation to describe the electric field distribution inside the metallized film capacitor. Specifically, the electric field strength is calculated through the gradient of the electric potential, and the electric field distribution satisfies the following Poisson equation: where φ is the electric potential distribution function, with the unit of volt, representing the electrostatic potential of the electric field; ∈ is the permittivity, with the unit of farad per meter, representing the dielectric property of the material; ρ is the charge density, with the unit of coulomb per cubic meter, which is determined by the applied voltage and the charge distribution in the dielectric.

[0038] As an option, in some embodiments, the boundary condition of the electric field model can be set such that a constant voltage V is applied to the outer electrode of the capacitor in , the inner electrode is grounded (φ = 0), and there is no net charge flow at the dielectric boundary, satisfying the condition that the derivative of the electric potential is zero.

[0039] In this embodiment, the thermal field model is based on the heat conduction equation to describe the thermal field distribution inside the capacitor. Specifically, the heat source caused by partial discharge and dielectric loss is expressed by the following formula: where Q is the heat source term, with the unit of watt per cubic meter, representing the local heat generation; ω is the electric field frequency, with the unit of radian per second; ∈″ is the dielectric loss factor, representing the dielectric loss property of the material; E is the electric field strength, with the unit of volt per meter.

[0040] Based on the heat source term, the temperature distribution of the capacitor satisfies the following heat conduction equation: where T is the temperature distribution function, with the unit of Kelvin; k is the thermal conductivity, with the unit of watt per meter per Kelvin; ρ is the material density, with the unit of kilogram per cubic meter; C p is the specific heat capacity, with the unit of joule per kilogram per Kelvin.

[0041] Specifically, the boundary condition of the thermal field model can be set such that heat dissipation occurs between the capacitor surface and the environment through natural convection, satisfying Newton's cooling law: -q·n = h(T - T ∞ ) where q is the heat flux density, with the unit of watt per square meter; h is the heat transfer coefficient, with the unit of watt per square meter per Kelvin; T∞ is the ambient temperature, in Kelvin.

[0042] In this embodiment, the force field model is used to describe the stress distribution inside the metallized film capacitor caused by thermal expansion and electric field force. The force field model satisfies the following mechanical equilibrium equation: where σ is the stress tensor, in Pascal, and f is the body force density, in Newton per cubic meter.

[0043] As an option, the body force density f includes two parts: thermal stress and electric field force: where α T is the thermal expansion coefficient, in per Kelvin, is the temperature gradient, in Kelvin per meter, ρ e is the charge density, in Coulomb per cubic meter, and E is the electric field strength vector, in Volt per meter.

[0044] In a possible implementation, the boundary condition of the force field model can be set as a fixed constraint at the edge of the capacitor, i.e., the displacement-free condition.

[0045] In this embodiment, a multi-physics field coupling model is constructed based on the above electric field, thermal field, and force field models. Generally, the solution of the coupling model is realized by the finite element method, and a commercial simulation software is used for discrete solution to obtain the distribution data of multi-physics fields inside the capacitor.

[0046] In some embodiments, to improve the calculation efficiency, the time series of multi-physics field signals can be processed in segments, and an iterative algorithm is used to solve the model. Specifically, the initial values of the electric field, thermal field, and force field are set through the initial conditions, and then the dynamic distributions of each physical field are iteratively solved until convergence.

[0047] Through the establishment and solution of the above coupling model, the multi-physics field distribution of the metallized film capacitor in the working state can be accurately described, providing a detailed physical basis for subsequent wavelet feature extraction and life prediction.

[0048] Through the above multi-physics field coupling modeling steps, the physical field distribution characteristics and their interactions inside the metallized film capacitor can be comprehensively analyzed, providing a physical basis and data support for feature extraction and life prediction in the subsequent steps.

[0049] In this embodiment, for step S3, after completing the multi-physical-field coupling modeling in step S2, in order to extract effective feature information from the complex multi-physical-field distribution, the present invention uses wavelet transform to perform time-frequency domain analysis on the multi-physical-field signals, and calculates the energy distribution characteristics of each physical-field signal through multi-scale decomposition. The multi-scale analysis of wavelet transform can effectively capture the local time-frequency characteristics of the signal and quantify the energy distribution of the signal at different scales, which provides high-dimensional feature data for subsequent health state inference and life prediction.

[0050] Generally, the frequency components of complex signals may vary with time, and it is difficult for traditional Fourier transform to characterize such dynamic characteristics. As an option, wavelet transform can characterize signal features in both the time domain and the frequency domain. Specifically, wavelet transform decomposes the signal into different time scales through wavelet basis functions, so as to extract the energy characteristics of the multi-scale signal. In the present invention, wavelet transform is mainly used to analyze the dynamic characteristics of electric field signals, thermal field signals and mechanical stress signals. The specific implementation content is as follows: In this embodiment, wavelet transform is used to perform multi-scale decomposition on the multi-physical-field signal x(t). Specifically, wavelet transform decomposes the signal through wavelet basis functions, and its mathematical expression is as follows: where W(a, b) is the wavelet coefficient, a is the scale factor, b is the time translation factor, is the wavelet basis function, and its expression is: where a is the scale factor, which is used to control the frequency resolution of the signal, b is the time translation factor, which is used to control the time resolution of the signal, and ψ(t) is the mother wavelet function, which satisfies the conditions of finite support and zero mean.

[0051] Generally, choosing a suitable wavelet basis function is crucial for signal characteristics. As an option, in this embodiment, Daubechies wavelet is selected as the wavelet basis function, which has good compact support and time-frequency localization characteristics and is suitable for analyzing complex dynamic characteristics in multi-physical-field signals.

[0052] After completing the wavelet transform, the energy characteristics of the signal at different time scales are extracted by calculating the energy of the wavelet coefficients. In this embodiment, the calculation method of wavelet energy is as follows: Specifically, for a certain scale a i , its corresponding wavelet coefficient is W(a i , b), then the total energy E i of the signal at this scale can be calculated by the following formula: Among them, E i represents the total energy of the signal at scale a i , with the unit being the integral value of the signal squared value. W(a i , b) is the time-frequency coefficient corresponding to the scale, indicating the contribution of the signal at scale a i and time b.

[0053] As an option, in actual implementation, the discrete wavelet transform is used to decompose the signal, and the energy is approximately calculated by the sum of the squares of the wavelet coefficients at discrete scales: Among them, b k represents the discrete time point, N is the total number of discrete sampling points of the signal. To normalize the energy distribution at different scales, the energy ratio P i of the signal at each scale is calculated: Among them, P i is the normalized energy ratio of the signal at the i-th scale, E i is the energy value of the signal at the i-th scale, N is the total number of decomposition scales, is the total wavelet energy of all scales, which is used for normalization processing.

[0054] In this embodiment, to quantify the complexity and chaos degree of the signal, the wavelet energy entropy is further calculated based on the multi-scale energy distribution. The wavelet energy entropy is calculated through the normalized energy ratio, and the expression is as follows: Among them, H is the wavelet energy entropy, P i is the normalized energy ratio of the signal at the i-th scale. Specifically, the larger the wavelet energy entropy value, the higher the complexity of the signal and the more significant the multi-scale characteristics; on the contrary, the signal tends to be more single or regular.

[0055] As an implementation method, the wavelet energy entropy can be calculated for each type of multi-physical field signal (such as electric field, thermal field, mechanical stress) respectively, and the calculation result is used as the feature input of the signal for subsequent health state assessment and life prediction.

[0056] In this embodiment, the number of scales N for signal decomposition needs to be selected according to the signal sampling rate and characteristic range. For example, for partial discharge signals, the frequency range is usually in the MHz level, and the decomposition scale can be set to 4 to 6 layers to cover the main frequency components and harmonic components.

[0057] In a possible implementation, when calculating the wavelet energy entropy, it can also be calculated separately for the multi-physical field signals in different time periods, and the dynamic entropy change of the signal is used as a time series feature and input into the health assessment model. This time-series processing method can further improve the accuracy of feature extraction.

[0058] Through the above feature extraction process of wavelet transform, the dynamic time-frequency characteristics of multi-physical field signals can be effectively captured, the energy distribution characteristics of signals at different scales can be quantified, and the complexity of signals can be evaluated through wavelet energy entropy. The extracted features can comprehensively characterize the operating state of multi-physical field signals, providing high-quality data support for the health state inference and life prediction in the subsequent steps.

[0059] In this embodiment, for step S4, after completing the wavelet transform and multi-scale energy distribution characteristic extraction in step S3, the present invention combines a dynamic Bayes network to dynamically infer the health state of the metallized film capacitor, so as to evaluate the variation law of its health state over time. Through the dynamic Bayes network, the characteristic information and time series relationship of multi-physical field signals can be used to perform real-time assessment and prediction of the health state, accurately depicting the aging and performance degradation characteristics of the capacitor during operation.

[0060] Generally, the dynamic Bayes network transforms complex multi-physical field features into a dynamic evolution description of the health state by constructing a state transition model and an observation model of the system. As an option, in the present invention, the state transition model describes the aging behavior of the capacitor based on a Poisson process, and the observation model is described by the correlation between the wavelet energy entropy and other features and the health state, so as to realize the dynamic inference of the health state. The specific implementation content is as follows: In this embodiment, the dynamic Bayes network is used to describe the time evolution law of the health state S t of the metallized film capacitor, as well as the correlation relationship between the observation variable O t and the health state.

[0061] Specifically, the dynamic Bayes network consists of the following two parts: State transition model: Describes the dynamic change of the system health state S t over time.

[0062] Observation model: Describes the relationship between the observation variable O t and the current health state S t Generally, the health state S t is an unobservable hidden variable directly, and its dynamic change is inferred through the state transition model; the observation variable O t ​Obtained from the extraction of multi-physical field signal features, including wavelet energy entropy, normalized energy distribution characteristics, and other related features.

[0063] In this embodiment, the health state S t The dynamic change of is assumed to follow a non-homogeneous Poisson process, and its state transition intensity λ(t) decays with time. The specific expression is as follows: λ(t) = λ 0 e -αt where λ(t) is the state transition intensity at time t, λ 0 is the initial state transition intensity, α is the aging rate coefficient, representing the decay speed of the health state, and t is the time.

[0064] The transition probability of the health state satisfies the following formula: P(S t |S t-1 ) = f(λ(t), Δt) where S t and S t-1 represent the health states at the current time and the previous time respectively, and f(λ(t), Δt) is the probability distribution function related to the state transition intensity and the time interval Δt.

[0065] As a possible implementation, the state transition model can be modeled using a Markov chain, that is, assuming that the current health state is only related to the state at the previous time, thereby simplifying the inference process.

[0066] In this embodiment, the observation variable O t includes features such as wavelet energy entropy H t , multi-scale normalized energy distribution P i (t), etc. These observation variables are associated with the health state S t through the following conditional probability distribution: P(O t |S t ) = g(S t ) where P(O t |S t ) is the conditional probability distribution between the observation variable and the health state, and g(S t ) is the function describing the feature distribution, and the specific form is selected according to the actual features.

[0067] As an option, the wavelet energy entropy H t can be directly used as a characterization index of the health state, and its relationship with the health state satisfies the following approximate linear relationship: H t = H 0 - βS t Among them, H t is the wavelet energy entropy at time t, H 0 is the initial energy entropy, and β is the influence factor of the health state on the entropy value change.

[0068] In this embodiment, the inference process of the dynamic Bayes network includes the following key links: Initial condition setting: The initial observation variable O extracted through step S3 0 Initializes the health state S 0 ; State transition inference: According to the state transition model P(S t ∣S t-1 ), the health state at the next moment is inferred; Observation update: According to the observation model P(O t ∣S t ), combined with the observation variable O at the current moment t The health state is corrected; Repeat the above steps until the health state of the complete time series is inferred.

[0069] As a possible implementation method, the Markov chain Monte Carlo method can be used for dynamic inference, that is, the state distribution is iteratively calculated through sampling technology to ensure the accuracy of the inference.

[0070] In this embodiment, the state transition intensity parameter λ 0 and the aging rate coefficient α need to be fitted through historical data. Specifically, based on the known health state evolution sequence, these parameters can be optimized by minimizing the prediction error.

[0071] As an extension, a time weighting mechanism can also be introduced in the dynamic Bayes network, so that recent observation variables have a higher weight on the health state inference, thereby enhancing the sensitivity of the model to dynamic changes.

[0072] In this embodiment, the health state S of the metallized film capacitor is dynamically inferred through the dynamic Bayes network t to obtain a complete health state change sequence. According to this sequence, the current health state and aging speed of the capacitor can be evaluated, and inputs for subsequent life prediction can be provided.

[0073] As an application, a health assessment report can be generated in combination with the health state sequence, including the current health level, aging trend, and suggestions on whether maintenance is required.

[0074] Through the above-mentioned inference process of the dynamic Bayes network, the present invention can realize the real-time assessment of the health state of the metallized film capacitor, providing strong support for the accurate diagnosis of the capacitor operation state and the life prediction.

[0075] In this embodiment, for step S5, after the dynamic inference of the health state in step S4 is completed, in order to further quantify the life change of the metallized film capacitor, the present invention establishes a life loss model based on the time evolution law of the health state and combines the characteristic change trend of the multi-physical field signals, and finally realizes the accurate prediction of the remaining life. Through the life loss model, the health state sequence can be directly associated with the remaining life, and combined with the health assessment report, maintenance suggestions for the capacitor can be provided.

[0076] Generally, the life of the metallized film capacitor is affected by multiple factors such as partial discharge, thermal loss, and mechanical stress. The life loss model quantifies the degradation rate of the health state and converts the cumulative effect of multiple factors into the time decay characteristic of the life. As an option, in the present invention, the life loss model adopts a mathematical description based on the cumulative change amount of energy entropy, and combines the dynamic health state inference result to calculate the remaining life of the capacitor. The specific implementation content is as follows: In this embodiment, the life loss model is used to describe the cumulative decay process of the capacitor's health state over time and quantify the impact of the health state change on the life. Specifically, the life loss model assumes that the remaining life L(t) is related to the initial life L 0 and the cumulative change amount of energy entropy, and its mathematical expression is: where L(t) is the remaining life at time t, in hours, L 0 is the initial life of the capacitor, in hours, determined by the manufacturer or historical test data, k is the life loss coefficient, indicating the sensitivity of the health state change to the life decay, H couple (τ) is the coupled entropy value at time (τ), indicating the complexity and interaction degree of the multi-physical field signals, represents the cumulative change amount of energy entropy.

[0077] Generally, the coupled entropy H couple (τ) can be calculated through the joint distribution of the wavelet energy entropy H(τ) extracted in step S3 and other eigenvalue. As a possible implementation, the wavelet energy entropy can be directly used as an approximation of the coupled entropy to simplify the calculation process.

[0078] In this embodiment, in order to improve the accuracy of the life loss model, the parameters k and L 0 in the model need to be optimized. Specifically, the optimization process can be based on the following steps: Collect the life test data of multiple sample capacitors, including the initial life, the sequence of health states during operation, and the actual failure time; Optimize the value of k by minimizing the error between the actual life and the predicted life; According to the experimental data under different working conditions, classify L 0 to ensure that the model is applicable to various working conditions.

[0079] As an extension, machine learning algorithms (such as linear regression or neural networks) can be combined to further fit the life loss model, improving the adaptability and accuracy of the model.

[0080] In this embodiment, according to the life loss model and the sequence of health states obtained in step S4, calculate the remaining life L of the capacitor remain . The specific calculation method is as follows: First, obtain the coupling entropy value H couple (τ) at different times through the dynamic inference result of the health state.

[0081] Then, perform a time integral on the coupling entropy to calculate the cumulative change: Finally, substitute the cumulative change into the life loss model to calculate the remaining life: In a possible implementation, to improve the calculation efficiency, numerical integration methods (such as the trapezoidal integration method or the Simpson integration method) can be used to approximately calculate the cumulative change of the coupling entropy. For real-time calculation requirements, the sliding window technique can be further combined to dynamically update the change of the coupling entropy.

[0082] In this embodiment, based on the calculation result of the remaining life, generate a health assessment report to intuitively reflect the operating state and maintenance requirements of the capacitor. The health assessment report mainly includes the following contents: Current health state: the latest value of the health state S obtained through step S4 t ; Remaining life prediction: L calculated according to the life loss model remain ; Maintenance suggestions: Judge whether preventive maintenance is required according to the threshold of the remaining life.

[0083] Generally, when the remaining life L remain is less than the preset threshold (such as 10% of the initial life), the system generates a maintenance reminder signal and provides a recommended maintenance time window.

[0084] As an option, the health assessment report can also display the time-varying trend of the remaining life and the historical evolution trajectory of the health status through a graphical interface, providing more intuitive decision-making support for users.

[0085] In this embodiment, the life loss model can be extended and applied to different types of metallized film capacitors, meeting the life prediction requirements under various working conditions. As an extension, the life loss model can be combined with a cloud data platform to achieve centralized monitoring and life prediction of multiple devices through real-time analysis of a large amount of capacitor data.

[0086] Through the above life loss modeling and remaining life calculation process, the present invention can accurately evaluate the operating state and remaining life of metallized film capacitors, discover potential failure risks in advance, and provide a scientific basis for the preventive maintenance and reliable operation of capacitors.

[0087] The present invention proposes a method for monitoring the life of a metallized film capacitor. By collecting electric field, thermal field, mechanical stress field, and partial discharge signals in real time, a multi-physical field coupling model is constructed to comprehensively analyze the complex physical behaviors inside the capacitor. Wavelet transform is used to extract features from the multi-physical field signals, calculate the multi-scale energy distribution characteristics and the complexity of the signals, and combine with a dynamic Bayes network to dynamically infer the health status of the capacitor, evaluate the change law of the health status in real time. Based on the evolution trend of the health status, a life loss model is established to calculate the remaining life of the capacitor and generate a health assessment report. This method overcomes the deficiencies of single-physical field monitoring, static evaluation, and empirical formula prediction in the prior art, realizes real-time monitoring, accurate evaluation, and personalized life prediction of metallized film capacitors, and provides a scientific basis for the preventive maintenance of equipment.

[0088] Although the embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for monitoring the life of a metallized film capacitor, characterized in that: The following steps are involved: The multi-physical field signals of the metallized film capacitor are collected in real time by using sensors, and the collected multi-physical field signals are pre-processed; A coupling model is established based on multi-physics fields to obtain the multi-physics field distribution of metallized film capacitors; The multi-physics field distribution is characterized by wavelet transform, and the multi-scale energy distribution characteristics of the signal are calculated; Based on the multi-scale energy distribution characteristics, the dynamic Bayes network is combined to dynamically reason about the health status of the capacitor and evaluate the changing law of its health status; Based on the law of health status changes, a life loss model is established to calculate the remaining life of metallized film capacitors and generate a health assessment report.

2. The method for monitoring the life of a metallized film capacitor according to claim 1, characterized in that: The method of collecting the multi-physical field signal of the metallized film capacitor comprises the following steps: The electric field signal is measured by an electric field probe, and the electric field strength is calculated by an electric field distribution model; The thermal field signal is measured by thermocouple or infrared thermal imager, and the temperature distribution is calculated by heat conduction model; The mechanical stress signal is measured by a piezoelectric sensor, and the stress distribution is calculated by a thermal stress and electric field force coupling model; The partial discharge signal is acquired through a high-speed data acquisition device with a sampling rate of no less than 1 GHz.

3. The method for monitoring the life of a metallized film capacitor according to claim 1, characterized in that: The coupling model is established based on the following relationship: The electric field distribution is described by Poisson’s equation; The thermal field distribution is described by the heat conduction equation, taking into account the heat source caused by partial discharge; The force field distribution is described by the interaction of thermal stress and electric field force.

4. The method for monitoring the life of a metallized film capacitor according to claim 1, characterized in that: The preprocessing of the multi-physics field signal includes: Use low-pass filtering to remove high-frequency noise; The electric field signal, thermal field signal and mechanical stress signal are normalized respectively; The time alignment method is used to synchronize the signals and ensure the time consistency of multi-physics field signals.

5. The method for monitoring the life of a metallized film capacitor according to claim 1, characterized in that: The wavelet transform adopts continuous wavelet transform, and the signal is decomposed into multiple scales through wavelet basis functions to extract energy distribution characteristics on different time scales.

6. The method for monitoring the life of a metallized film capacitor according to claim 1, characterized in that: The multi-scale energy distribution characteristics are used to calculate the energy entropy of multi-physical field signals. The energy entropy is calculated as follows: The collected multi-physics field signals are decomposed by wavelet, and the signal x(t) is decomposed into multiple scales using wavelet basis functions. The expression of wavelet transform is: Among them, W(a,b) is the wavelet coefficient, a is the scale factor, b is the time translation factor, is the wavelet basis function; The corresponding energy value is calculated for each scale wavelet coefficient obtained by wavelet decomposition. The calculation formula of the energy value is: HAVE BEEN i =|W(a i ,b)| 2 Among them, E i represents the wavelet energy value of the i-th scale; The wavelet energy values ​​of each scale are normalized to obtain the wavelet energy ratio of each scale. The normalization formula is: Among them, P i is the wavelet energy ratio of the i-th scale, E i is the energy value of the signal at the i-th scale, N is the total number of decomposition scales, is the sum of wavelet energies at all scales, used for normalization; Calculate the wavelet energy entropy. The expression of wavelet energy entropy is: Among them, H represents the wavelet energy entropy of the multi-physics field signal, reflecting the complexity and chaos degree of the signal; Wavelet energy entropy is used as a signal feature to input into the life prediction model to characterize the dynamic change characteristics of multi-physical field signals.

7. The method for monitoring the life of a metallized film capacitor according to claim 1, characterized in that: The construction of the dynamic Bayes network includes the following contents: Define the system health state and observed variables. The system health state follows the state transition probability distribution over time. The state transition probability is related to the discharge intensity and aging rate. The relationship between observed variables and health status is described by conditional probability distribution; Based on the real-time observed multi-physics field signal eigenvalues, a dynamic reasoning method is used to estimate the system health status.

8. The method for monitoring the life of a metallized film capacitor according to claim 1, characterized in that: The life loss model is established based on the cumulative change in energy entropy of multi-physical field signals. The larger the cumulative change, the faster the life decay rate.

9. The method for monitoring the life of a metallized film capacitor according to claim 1, characterized in that: The life loss model is described by the following relationship: The remaining life is the inverse function of the initial life and the cumulative change in energy entropy; When the cumulative change in energy entropy exceeds a preset threshold, the remaining life prediction value decreases significantly.

10. The method for monitoring the life of a metallized film capacitor according to claim 1, characterized in that: The health assessment report includes the current health status of the metallized film capacitor, the remaining life prediction result, and an early warning prompt as to whether maintenance is required. When the remaining life is lower than a set threshold, the system generates an alarm signal.

Images