A capacitor accelerated aging reliability analysis method and system based on censored data processing

Abstract

This application discloses a method and system for reliability analysis of accelerated aging of capacitors based on censored data processing. The method involves acquiring raw data from accelerated aging tests of capacitors and standardizing the data to obtain standardized sample data. Based on the standardized sample data, the data is grouped using test type, capacitor model, test temperature, applied surge voltage, and applied ripple current as operating condition keys. The samples under each operating condition are then sorted according to the test time. The number of non-failed and failed samples at each test time is statistically analyzed based on the sample status, constructing a grouped censored sample that includes interval censored samples and right-censored samples. A feature vector is constructed, and a Weibull regression model is used to characterize the capacitor lifetime distribution, where the scale parameter varies with the operating condition and the shape parameter is shared across different operating conditions. A negative log-likelihood objective function is established based on the grouped censored samples, and the negative log-likelihood objective function is optimized to obtain the parameters of the Weibull regression model. Based on the trained Weibull regression model, the reliability curve and lifetime index under the target operating condition are calculated. When the target operating condition is determined to be a fully right-censored condition, the reliability curve is right-shifted and then output. This application can simultaneously process interval censored data and right censored data in accelerated aging tests of capacitors, improving the accuracy and engineering applicability of reliability curve plotting and life assessment.

Description

Technical Field

[0001] This application belongs to the field of power electronic device health assessment and life prediction technology, specifically involving a method and system for accelerating aging reliability analysis of capacitors based on censored data processing. Background Technology

[0002] Capacitors are key basic components in power electronic equipment, widely used for DC bus support, filtering, energy storage, and ripple suppression. To obtain lifespan information in a short period of time, aging tests are usually conducted under accelerated stress conditions such as high temperature, high surge voltage, and high ripple current. Based on the test results, the reliability curve and key lifespan indicators under target operating conditions are deduced.

[0003] Existing accelerated aging reliability analysis methods typically suffer from the following shortcomings in engineering applications: First, on-site testing often employs sampling or periodic recording at fixed time points to obtain sample status, resulting in censored or right-censored data in a grouped statistical sense rather than precise failure moments. Second, for some operating conditions, failure events do not occur before the observation deadline; if ordinary Weibull curves are still directly output, reliability will decline prematurely, inconsistent with experimental facts. Third, existing analysis processes typically revolve around fields such as test type, capacitor model, test temperature, applied surge voltage, and applied ripple current, but lack a unified and standardized modeling and processing mechanism for censored data under different operating conditions. Fourth, existing processes often lack standardized definitions for handling missing data, easily leading to sample bias, grouping errors, or parameter distortion.

[0004] In some engineering practices, experimental data may be simplified to facilitate model verification or local analysis, retaining only key fields directly related to modeling. However, in actual software deployments and formal patent protection scenarios, it is still necessary to retain some of the deleted fields to enable more refined operating condition classification and more robust lifespan analysis in situations involving multiple manufacturers, multiple rated voltage levels, and multiple package sizes. Summary of the Invention

[0005] The purpose of this application is to provide a method and system for accelerating the reliability analysis of capacitors based on censored data processing, so as to solve the problems in the prior art of difficulty in standardizing the processing of interval censored and right-censored data, difficulty in robustly estimating Weibull distribution parameters, and inconsistency between the reliability curve and experimental facts under the full right-censored condition.

[0006] To achieve the above objectives, this application adopts the following technical solution: In a first aspect, this application provides a method for accelerating the reliability analysis of capacitors based on censored data processing, including: Obtain raw data from accelerated aging tests of capacitors, and standardize the raw data to obtain standardized sample data; the raw data includes at least the following operating condition data items: test type, capacitor model, test temperature, surge voltage applied during the test, and ripple current applied during the test, as well as observation data items such as test time and sample status; Based on the standardized sample data, the test type, capacitor model, test temperature, surge voltage applied during the test, and ripple current applied during the test are used as operating condition keys to group the samples. The samples under each operating condition are sorted according to the test time. The number of non-failed samples and the number of failed samples at each detection time are counted in combination with the sample status. A grouped censored sample containing interval censored samples and right censored samples is constructed. We construct feature vectors and use a Weibull regression model to characterize the capacitor lifetime distribution, where the scale parameter varies with operating conditions and the shape parameter is shared across different operating conditions. A negative log-likelihood objective function is established based on the grouped censored samples, wherein the probability contribution is determined based on the interval censoring probability of the interval censored samples and the survival probability of the right censored samples; The negative log-likelihood objective function is optimized and solved to estimate the parameters of the Weibull regression model, which include at least the shared shape parameter. Based on the trained Weibull regression model, the reliability curve and lifetime index under the target operating condition are calculated. When the target operating condition is determined to be a fully right-censored operating condition, the reliability curve is corrected by a right shift and then output.

[0007] As a further improvement to this application, the step of constructing a grouped censored sample containing interval censored samples and right censored samples based on the standardized sample data, grouping according to the working condition data items, and statistically analyzing the observation data items includes: Using the test type, capacitor model, test temperature, applied surge voltage, and applied ripple current as operating condition keys, the standardized sample data is grouped to obtain the time series corresponding to each operating condition; The samples under each working condition are sorted by time, and the time interval in which the failed sample falls is determined based on the change in the number of non-failed samples between adjacent detection times. Samples whose failure time falls between two adjacent detection times are recorded as interval censored samples. Samples for which no failure was observed up to the last detection time of a single operating condition are recorded as right-censored samples; When the total number of failures in a single operating condition is zero, the operating condition is marked as a fully right-censored operating condition.

[0008] As a further improvement to this application, the construction of the feature vector includes: The feature vector is constructed by combining test type dummy variables, capacitor model dummy variables, and continuous stress characteristics. The continuous stress characteristics include at least the standardized test temperature characteristics, the surge voltage characteristics applied during the test, and the ripple current characteristics applied during the test.

[0009] As a further improvement to this application, the Weibull regression model satisfies:

[0010]

[0011] in, The scale parameters corresponding to the target working condition. Shape parameters to be shared across different operating conditions; The continuous stress characteristics include at least temperature normalization characteristics. Surge voltage normalization characteristics and ripple current normalization characteristics ; The temperature normalization feature satisfies ; The surge voltage normalization characteristic satisfies ; The ripple current normalization characteristic satisfies .

[0012] in, For the test temperature, The surge voltage applied to the test, The ripple current applied to the test, , and These are the preset reference temperature, reference surge voltage, and reference ripple current, respectively.

[0013] As a further improvement of this application, the total objective function of the negative log-likelihood objective function is formed by accumulating the interval censored likelihood term and the right censored likelihood term for each working condition and time interval, and then adding a parameter regularization term. The interval censoring likelihood term is constructed based on the condition failure probability and the number of failure samples within the corresponding time interval. The condition failure probability is determined by the survival function values ​​of the Weibull regression model at the current time node and the previous time node. The right-censored likelihood term is constructed based on the survival probability at the cutoff time.

[0014] As a further improvement to this application, the optimization of the negative log-likelihood objective function to estimate the parameters of the Weibull regression model includes: The negative log-likelihood objective function is iteratively optimized using multiple sets of different initial values ​​for the shared shape parameters. From the results of multiple optimizations, the set of parameters that minimizes the total objective function value is selected as the final estimated model parameters.

[0015] As a further improvement to this application, the standardization processing of the original data includes processing the identified missing data according to whether the missing field belongs to a preset key field using different rules; The key fields are those that participate in the grouping statistics and feature vector construction.

[0016] As a further improvement to this application, when the missing field belongs to a preset key field, the rules adopted include: removing the corresponding sample, retrospectively recording based on experimental data, or restoring the field value based on the fixed attributes of samples in the same batch.

[0017] As a further improvement to this application, when the missing field does not belong to the preset key field, the rules adopted include: using the mode of the same working condition to fill in the missing field, the median of the same working condition to fill in the missing field, the interpolation of the nearest time to fill in the missing field, or the preset default value to fill in the missing field, and generating the corresponding missing identification variable.

[0018] When key fields are missing in the original data, the missing fields are recovered based on external data sources, and the recovered fields are used in subsequent processing.

[0019] As a further improvement to this application, the external data source is at least one of the following: test data, sample master data table, product specification sheet, equipment configuration file, or fixed attributes of samples in the same batch.

[0020] As a further improvement to this application, the step of performing a right shift correction on the reliability curve and then outputting it when the target operating condition is determined to be a fully right-censored operating condition includes: When the target condition is marked as the fully right-censored condition, a curve start-up time greater than zero is determined based on the cutoff observation time of the condition and the preset correction rule. Before the curve shift time, output a curve segment with a reliability of 1; and after the curve shift time, output a reliability curve segment calculated based on the Weibull regression model.

[0021] As a further improvement to this application, the calculation of the life index under the target operating condition includes: B10 lifespan: ; Median lifespan: ; in, The dimensional parameters under the target working condition are... The shared shape parameter.

[0022] As a further improvement to this application, the standardization processing of the original data further includes: Fields with inconsistent names but equivalent semantics in the original data are uniformly mapped to standard field names through a preset field alias mapping table; Convert the category field in the original data into a categorical variable; The original test times in the raw data are uniformly converted into values ​​on a preset time scale.

[0023] The operating condition data items specifically include: The test includes multiple fields such as test type, capacitor model, test temperature, applied surge voltage, and applied ripple current.

[0024] Secondly, this application provides a capacitor accelerated aging reliability analysis system based on censored data processing, comprising: The data preprocessing module is used to acquire the raw data of the accelerated aging test of the capacitor and to standardize the raw data to obtain standardized sample data; the raw data includes at least the operating condition data item. The grouped censoring construction module is used to perform grouped statistics and sorting according to the working condition data items based on the standardized sample data, and construct a grouped censored sample that includes interval censored samples and right censored samples; The model building module is used to construct feature vectors and uses a Weibull regression model with scale parameters varying with operating conditions and shape parameters shared across different operating conditions to characterize the capacitor lifetime distribution. The parameter estimation module is used to establish a negative log-likelihood objective function based on the grouped censored samples, and to optimize and solve the negative log-likelihood objective function to estimate the parameters of the Weibull regression model, wherein the parameters include at least the shared shape parameter; wherein the probability contribution is determined based on the interval censoring probability of the interval censored samples and the survival probability of the right censored samples; The prediction and correction module is used to calculate the reliability curve and lifetime index under the target operating condition based on the trained Weibull regression model, and when the target operating condition is determined to be a fully right-censored operating condition, the reliability curve is corrected by right shift and then output.

[0025] Compared with the prior art, the beneficial effects of this application are as follows: This application, by grouping and constructing features based on test type, capacitor model, test temperature, applied surge voltage, and applied ripple current, can comprehensively characterize test conditions and improve the distinguishability between different sample groups. By uniformly handling missing key fields, missing auxiliary fields, and inconsistent field names, the interference of differences in the quality of original test records on modeling results can be reduced, thereby enhancing the reproducibility of data governance and the robustness of model training. By transforming grouped statistical data into grouped censored samples and constructing conditional failure probabilities at the interval level, interval censored data formed under fixed detection intervals can be directly processed, while also being compatible with right-censored data. By constructing shared shape parameters… Operating condition related dimensional parameters The Weibull regression model can integrate test type, capacitor type, and continuous stress characteristics while maintaining model compactness, achieving reliability modeling more suitable for engineering scenarios. By identifying the fully right-censored operating condition and setting a curve right-shift correction mechanism, the output curve is made to better match the experimental facts near the observation cutoff time, facilitating conservative lifetime interpretation by engineers.

[0026] Furthermore, by providing a visual interactive output of reliability curves, B10 lifetime, and median lifetime, researchers can directly conduct lifetime assessments, scheme comparisons, and experimental design optimizations under different operating conditions. Attached Figure Description

[0027] Figure 1 This is a flowchart illustrating the overall process of a capacitor accelerated aging reliability analysis method based on censored data processing, as described in this application. Figure 2 This is a schematic diagram illustrating the construction of grouped censored samples for this application; Figure 3 This is a schematic diagram of the Weibull regression modeling for the shared shape parameters in this application; Figure 4 This is a schematic diagram illustrating the rightward shift correction of the reliability curve under the fully right-censored operating condition of this application. Detailed Implementation

[0028] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the protection scope of this application.

[0029] This application provides a capacitor accelerated aging reliability analysis method based on censored data processing, capable of performing normalized missing data processing on capacitor accelerated aging test data, simultaneously processing grouped interval censored data and right-censored data, and outputting reliability curves, B10 lifetime, and median lifetime. This addresses the problems in existing technologies such as difficulty in normalizing missing data processing, difficulty in simultaneously integrating interval censoring and right-censoring information, and inconsistencies between reliability curves and experimental data under full right-censoring conditions. The specific solution is as follows: Obtain raw test data containing fields such as test type, capacitor model, test temperature, applied surge voltage, applied ripple current, test time, and status; perform field mapping, missing data processing, and standardization on this data; construct grouped censored samples based on the operating condition fields and test time; construct a Weibull regression model with shared shape parameters, allowing the scale parameter to vary with the operating condition; establish a negative log-likelihood objective function based on the interval condition failure probability, and estimate the model parameters using a multi-initial-value iterative optimization method; output the reliability curve, scale parameter, shared shape parameter, B10 lifetime, and median lifetime after receiving the target operating condition; for fully right-censored operating conditions that meet preset matching conditions with the training operating conditions and have no failure events, perform a right-shift correction on the curve.

[0030] Specifically, such as Figure 1 As shown, this application provides a method for accelerated aging reliability analysis of capacitors based on censored data processing, including the following steps: S1. Obtain the raw data of the accelerated aging test of the capacitor, and perform field name mapping, field integrity verification, missing data processing, categorization processing and time scale unification on the raw data to obtain standardized sample data; the raw data shall include at least the test type, capacitor model, test temperature, surge voltage applied in the test and ripple current applied in the test and other operating condition data items, as well as test time and sample status observation data items. Specifically, step S1 includes: Construct a field alias mapping table to uniformly map fields with inconsistent names but equivalent semantics to preset standard field names; The test type, capacitor model, test temperature, surge voltage and ripple current applied during the test used to define the operating condition key, the test time used for time sorting and censoring interval division, and the sample status used for sample health status identification are defined as key data items. When a key field is missing, the corresponding sample is removed, the experimental data is retrospectively entered, the sample master data table is restored, or the field value is restored based on the fixed attributes of the same batch. For non-critical auxiliary fields that do not participate in the current round of grouping modeling, the missing data are filled in using the same working condition mode, median, nearby time interpolation, or preset default value, and a missing data identifier variable is generated. The category fields in the test type and capacitor model are converted into categorical variables, and the original test time is uniformly converted into an hour scale according to the preset time scaling factor. The sample state is used to characterize the observation state of the sample at the corresponding detection time, including at least the non-failed state and the failed state; the right-censored sample is determined by the sample that has not been observed to be failed up to the last detection time.

[0031] S2. Using the working condition data item and the test time as the working condition-time joint statistical key, perform the first grouping statistics on the standardized sample data to obtain the number of observed samples and the number of non-failed samples for each working condition at each test time; then use the working condition data item as the working condition key to perform the second grouping to form the time series corresponding to each working condition, and construct the grouped censored samples divided by working condition. The operating condition field includes test type, capacitor type, test temperature, surge voltage applied during the test, and ripple current applied during the test; Among them, the failure samples fall into the corresponding time interval based on the change in the number of surviving samples between adjacent detection times. Samples whose failure time is between the previous detection time and the next detection time are recorded as interval censored samples. Samples that have not failed by the last detection time are recorded as right censored samples. And it is identified whether each working condition belongs to the fully right censored working condition. Step S2 includes: Using the aforementioned operating condition data items and the test time at a unified time scale as the joint statistical key of operating condition and time, the first grouping was performed to obtain the number of observation samples for each operating condition at each detection time node. and the number of non-failed samples ; Then, the operating condition data items are used as operating condition keys for a second grouping to form a time series corresponding to a single operating condition; The time series for a single operating condition are arranged in ascending order of test time, and the... The sequence is processed to accumulate minimum values ​​so that the number of non-failed samples monotonically does not increase with time. Let the initial sample size for a single working condition be denoted as . The decrease in the number of non-failed samples between adjacent detection times is recorded as the number of failed samples in that time interval, to characterize the interval censored samples that fail between the previous detection time and the current detection time; Samples for which no failure was observed up to the last detection time of a single operating condition are recorded as right-censored samples; Initial sample size for a single working condition The difference between the number of non-failed samples at the end of the time step is taken as the total number of failures. When the total number of failures is zero, the operating condition is marked as a fully right-censored operating condition, and the maximum observation time of this operating condition is marked as the cutoff time. .

[0032] S3. Construct a feature vector containing dummy variables for test type and capacitor model, as well as continuous features of test temperature, applied surge voltage and applied ripple current. Use a Weibull regression model with scale parameters varying with operating conditions and shape parameters shared across different operating conditions to characterize capacitor lifetime distribution. In S3, the feature vector It consists of an intercept term, a test type dummy variable, a capacitor model dummy variable, and continuous stress characteristics; wherein, the continuous stress characteristics include at least the test temperature characteristics, the surge voltage characteristics applied during the test, and the ripple current characteristics applied during the test.

[0033] The Weibull regression model satisfies:

[0034]

[0035] in, The scale parameters corresponding to the target working condition. Shape parameters to be shared across different operating conditions; The continuous stress characteristics include at least temperature normalization characteristics. Surge voltage normalization characteristics and ripple current normalization characteristics ; The temperature normalization feature satisfies ; The surge voltage normalization characteristic satisfies ; The ripple current normalization characteristic satisfies .

[0036] in, For the test temperature, The surge voltage applied to the test, The ripple current applied to the test, , and These are the preset reference temperature, reference surge voltage, and reference ripple current, respectively.

[0037] S4. Establish a negative log-likelihood objective function based on the grouped censored samples; For the interval censored samples, the likelihood term is constructed using the interval probability that the failure time falls into the corresponding time interval; For the right-censored samples, the likelihood term is constructed using the survival probability of the samples that are still valid at the cutoff observation time; The likelihood terms under different working conditions are accumulated, and a parameter regularization term is added to form the overall objective function of the Weibull regression model. In step S4, a negative log-likelihood objective function is constructed based on the interval censored samples and the right censored samples, specifically including: For the i-th working condition and the j-th time interval, let the previous time node be . The current time point is The probability of condition failure within that time interval is... satisfy: ; in, and These are the survival function values ​​of the i-th working condition at the current time node and the previous time node, respectively; Located in the time interval The number of samples that failed within the period is denoted as The number of risk set samples at the beginning of this time interval is denoted as Then the likelihood term corresponding to the censored samples in the interval is... and Build; The remaining samples that have not been observed to be invalid after the last detection time will be right-censored samples, and their likelihood contribution will be constructed from the survival probability at the cutoff time, or equivalently incorporated into the survival term of the last time interval. The interval censored likelihood term and the right censored likelihood term for each time interval under each working condition are accumulated, and a parameter regularization term is added to obtain the total negative log-likelihood function:

[0038] in, The regularization coefficient is . These are the model parameters excluding the intercept term.

[0039] S5. Estimate the parameters of the Weibull regression model using a multi-initial-value iterative optimization method, obtain the model parameters and shared shape parameters, and save the model metadata; In step S5, a method is used for shared shape parameters. Multiple initial values ​​are iteratively optimized, and the set with the smallest objective function value is selected as the final model parameters from the multiple optimization results; wherein, the shared shape parameter The logarithmic parameter is guaranteed to be positive by using an exponential mapping.

[0040] S6. Receive the target operating condition input and calculate the reliability function, scale parameter, B10 lifetime and median lifetime under the target operating condition based on the trained Weibull regression model. In step S6, the scale parameters under the target working condition are used. and shared shape parameters Calculate lifetime indicators, which include: B10 lifespan: ; Median lifespan: ; The reliability curve is adaptively generated based on the maximum observation time of the training condition and the 99% failure quantile time of the target condition, and the reliability curve under the target condition is output. .

[0041] S7. When the target working condition and the training working condition meet the preset matching conditions and the corresponding working condition is a fully right-censored working condition, the reliability curve is corrected by right shift and then the reliability curve under the target working condition is output. The preset matching conditions include that the category field is consistent and the continuous stress field is within the preset tolerance range.

[0042] In step S7, the right shift correction includes: When the target condition and the training condition meet the preset matching conditions and the condition is a full right censoring condition, the product of the deadline and the preset right shift ratio is used as the curve shift start time. ; Set the reliability to 1 before the curve starts shifting; After the curve start-up time, according to

[0043] Output the corrected reliability curve; The curve start-up time is superimposed when calculating B10 lifetime and median lifetime. The preset right shift ratio is a constant determined based on historical experimental experience, data distribution characteristics, or model configuration parameters.

[0044] In a preferred embodiment, samples with missing key fields participating in grouping and modeling are removed, retrospectively supplemented, or their constant field values ​​are restored based on experimental data. For samples with missing auxiliary fields, statistical completion under the same working conditions, interpolation of adjacent times, or completion using preset values ​​are adopted, and missing identification variables are recorded to preserve sample utilization and improve the interpretability of subsequent analysis.

[0045] The present application will be further described below with reference to the accompanying drawings and embodiments. It should be understood that the following embodiments are for illustrative purposes only and are not intended to limit the scope of protection of the present application. For those skilled in the art, equivalent substitutions and modifications made without departing from the concept of the present application should fall within the scope of protection of the present application.

[0046] Example 1: Raw Experimental Data and Field System like Figure 1As shown, in this embodiment, the raw test data can be stored as a spreadsheet file, a database table, or a structured record exported by test management software. The raw fields include at least the test type (Exp), test time (Time), capacitor model (Cap_Model), test temperature (Tem), applied surge voltage (Surge_Voltage), applied ripple current (Ripple_current), and sample status (OK).

[0047] The test type, capacitor model, test temperature, applied surge voltage, and applied ripple current are used to define the test conditions; the test time is used to construct a time series, identify interval censoring and right censoring, and participate in the calculation of lifetime indicators; the sample status is used to characterize the observation status of the sample at the corresponding time node.

[0048] When some sample records have inconsistent field names, missing fields, or different formats, standardized sample data can be formed through field mapping, missing field handling, and time scaling.

[0049] For ease of understanding, the main variables and fields involved in this application are described in Table 1.

[0050] Table 1. Description of main variables and fields

[0051] Example 2: Field Mapping and Missing Data Processing The system first establishes a mapping relationship between field aliases. For example, when the original data contains different spellings such as Ripple_current, RippleCurrent, and Ripple_current, they are uniformly mapped to the standard field name Ripple_current.

[0052] The integrity of the fields is then verified. For key fields such as test type, capacitor model, test temperature, applied surge voltage, applied ripple current, test time, and status, if records are missing, one of the following rules is preferred: First, directly remove the sample; second, retrospectively record the missing data based on the test data; third, if the field is a global constant in the batch of samples, restore a unified value from the equipment configuration file.

[0053] For non-critical auxiliary fields, the following methods can be used to complete the data: mode completion under the same working conditions, median completion under the same working conditions, interpolation completion based on nearby time, or preset default value completion. At the same time, missing identification variables are generated so that the completed samples can be identified in subsequent quality tracking, model interpretation, or anomaly review.

[0054] After completing the field processing, the test type, capacitor model and other category fields are converted into categorical variables; the test time is uniformly converted into Time_h on an hourly scale according to the preset time scaling factor, so as to provide a unified time benchmark for subsequent grouping and modeling.

[0055] Example 3: Construction of Grouped Censored Samples In this embodiment, the standardized sample data are first grouped and statistically analyzed using the test type, capacitor model, test temperature, applied surge voltage, applied ripple current, and test time at a uniform time scale as joint keys to obtain the total number of observed samples at each time point. and the number of non-failed samples .

[0056] Subsequently, the test type, capacitor model, test temperature, applied surge voltage and applied ripple current were used as the test conditions for a second grouping, and the results of the first grouping were organized into time series corresponding to multiple test conditions.

[0057] For each working condition, The sequence performs cumulative minimum value processing to prevent the number of non-failed samples from increasing over time, thereby eliminating non-physical fluctuations caused by detection errors, data entry errors, or statistical disturbances.

[0058] For each working condition, take the initial sample size. and with The difference between the number of samples that did not fail at the last moment and the total number of failures is taken as the total number of failures; if the total number of failures is zero, the operating condition is determined to be a fully right-censored operating condition, and its maximum observation time is recorded as . The above processing procedure is as follows: Figure 2 As shown.

[0059] Example 4: Weibull Regression Model with Shared Shape Parameters To balance model expressiveness and parameter robustness, this application employs a Weibull regression model where the scale parameter varies with the operating condition, while the shape parameter is shared across different operating conditions. The reliability function for the target operating condition x is defined as: ,in For scale parameters, To share shape parameters.

[0060] In one specific implementation, the feature vector It consists of the following components: intercept term; test type dummy variable; capacitor type dummy variable; and temperature normalization feature. Surge voltage normalization characteristics and ripple current normalization characteristics .

[0061] Among them, the temperature normalization feature satisfies Surge voltage normalization characteristics satisfy The ripple current normalization characteristic satisfies .

[0062] The aforementioned feature construction method can express both the discrete differences between different test types and different capacitor models, and also characterize the continuous influence of accelerated stresses such as temperature, surge voltage, and ripple current on lifetime-scale parameters. Its modeling relationship is as follows: Figure 3 As shown.

[0063] Example 5: Grouping and Censoring of Negative Log-Likelihood Construction For the i-th working condition, let its detection time series be... , , ..., Let the number of risk set samples at the beginning of the j-th time interval be... The number of failures in the corresponding interval is Based on the Weibull reliability function, the conditional failure probability for this interval can be obtained. .

[0064] The interval likelihood of this working condition can then be expressed as a binomial structure. By combining the interval likelihoods of all working conditions, taking the negative logarithm, and introducing a regularization term, the optimization objective can be obtained: .

[0065] in, The regularization coefficient is . These are the model parameters excluding the intercept term.

[0066] This objective function can handle both grouped interval censored data formed by fixed detection intervals and right censored conditions that have not failed before the deadline, making it more suitable for the statistical characteristics of real accelerated aging test records.

[0067] Example 6: Model Parameter Training and Metadata Storage Since the objective function of a Weibull regression model with shared shape parameters typically lacks an analytical solution, this embodiment employs a derivative-free iterative optimization method with multiple initial values ​​to train the model parameters. Specifically, the scaling parameters are first initialized based on the median of the trial time for all samples. Then, multiple sets of starting points are set for the logarithmic values ​​of the shared shape parameters, and iterative solutions are performed for each set. Finally, the set with the smallest objective function value is selected as the final parameters.

[0068] After training is complete, the system saves the model parameters. Shared shape parameters The model metadata includes at least a set of test types, a set of capacitor models, training boundaries for temperature / surge voltage / ripple current, time scaling factor, and full right censoring correction ratio, for subsequent prediction and visualization output.

[0069] Example 7: Reliability Curve Prediction and Lifetime Index Output During the prediction phase, users input target operating conditions through a graphical user interface, including test type, capacitor type, test temperature, applied surge voltage, and applied ripple current. The system generates feature vectors based on the input operating conditions. Calculate scale parameters and shared shape parameters and based on Output the reliability curve under the target operating condition.

[0070] At the same time, the system calculates key lifespan indicators: , .in, The median lifetime corresponds to a reliability of 90%. The corresponding lifespan when the reliability is 50%.

[0071] To ensure that the display range covers both the upper limit of the training samples and the main failure intervals, the system preferably adaptively determines the plotting endpoint based on the maximum training sample time and the 99th percentile time of failure, and displays it synchronously on the interface. , , , And training range prompts.

[0072] Example 8: Right Shift Correction for Full Right Deletion Case When the operating condition to be predicted is completely identical to a certain operating condition in the training set, and that operating condition is identified as a fully right-censored operating condition, if the standard Weibull reliability function is still used directly, the curve will immediately drop after t=0, which is inconsistent with the experimental fact that "no failure was observed before the deadline". Therefore, this application proposes a right-shift correction strategy.

[0073] Let the deadline for this working condition be... The right shift ratio is The curve start-up time is defined as: Preferably, Take 0.80.

[0074] The revised reliability curve is defined as: when hour, ;when hour, Accordingly, the corrected B10 lifetime and median lifetime are as follows: , . Figure 4 The right-shift correction method is illustrated.

[0075] This application also provides a capacitor accelerated aging reliability analysis system based on censored data processing. Based on the above-described capacitor accelerated aging test analysis method, the system includes a data preprocessing module, a grouped censoring construction module, a model construction module, a parameter estimation module, and a prediction and correction module. The data preprocessing module is used to acquire raw data from the capacitor accelerated aging test and standardize the raw data to obtain standardized sample data; the raw data includes at least the test type, test time, capacitor model, test temperature, applied surge voltage, applied ripple current, and status field. The grouped censoring construction module is used to perform grouped statistics and sorting based on the standardized sample data according to the working condition field and combined with the test time, and construct a grouped censored sample that includes interval censored samples and right censored samples; The model building module is used to construct feature vectors and uses a Weibull regression model with scale parameters varying with operating conditions and shape parameters shared across different operating conditions to characterize the capacitor lifetime distribution. The parameter estimation module is used to establish a negative log-likelihood objective function based on the grouped censored samples, and to optimize and solve the negative log-likelihood objective function to estimate the parameters of the Weibull regression model, wherein the parameters include at least the shared shape parameter; wherein the probability contribution is determined based on the interval censoring probability of the interval censored samples and the survival probability of the right censored samples; The prediction and correction module is used to calculate the reliability curve and lifetime index under the target operating condition based on the trained Weibull regression model, and when the target operating condition is determined to be a fully right-censored operating condition, the reliability curve is corrected by right shift and then output.

[0076] The data preprocessing module is used to complete field mapping, missing data cleaning, categorization, and time scale unification; the prediction and correction module is used to receive target operating condition input and display curves and key information.

[0077] Table 2 System Module Function Description

[0078] This application is applicable to offline analysis tools implemented with engineering software such as MATLAB, as well as embedded test management platforms, laboratory life management systems, and online reliability assessment systems.

[0079] This capacitor accelerated aging reliability analysis system acquires accelerated aging test data including fields such as test type, capacitor model, test temperature, applied surge voltage, applied ripple current, test time, and status. It then performs field mapping, field validation, missing data processing, and standardization on the data. The system groups and statistically analyzes the data according to the operating condition fields and test time, constructing grouped censored samples and identifying interval censored samples, right-censored samples, and fully right-censored operating conditions. It establishes a Weibull regression model where scale parameters change with operating conditions, and shape parameters are shared across different operating conditions. Based on interval censoring probability and right-censoring survival probability, it constructs a negative log-likelihood function and completes parameter estimation. Upon receiving the target operating condition, it outputs the corresponding reliability curve, scale parameters, shared shape parameters, B10 lifetime, and median lifetime. When the target operating condition and training operating condition meet preset matching conditions and are fully right-censored, the reliability curve is right-shifted.

[0080] This application can simultaneously process interval censored data and right censored data in accelerated aging tests of capacitors, improving the accuracy and engineering applicability of reliability curve plotting and life assessment.

[0081] A third aspect of this application is to provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the aforementioned accelerated aging test analysis method for capacitors. The device also includes a communication interface and a bus.

[0082] A fourth aspect of this application is to provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method for analyzing accelerated aging tests of capacitors.

[0083] A fifth aspect of this application is to provide a computer program product, which includes computer instructions that instruct a computer to execute the above-described capacitor accelerated aging test analysis method.

[0084] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0085] These computer program instructions can also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0086] This application may take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application may take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, readable storage media, optical storage, etc.) containing computer-usable program code.

[0087] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0088] Obviously, the described embodiments are only some, not all, of the embodiments in this application. All other embodiments obtained by those skilled in the art based on the embodiments in this application without inventive effort should fall within the scope of protection of this application.

[0089] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application and not to limit them. Although this application has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation methods of this application. Any modifications or equivalent substitutions that do not depart from the spirit and scope of this application should be covered within the protection scope of this application.

Claims

1. A method for accelerating the reliability analysis of capacitors based on censored data processing, characterized in that, include: Obtain raw data from accelerated aging tests of capacitors, and standardize the raw data to obtain standardized sample data; the raw data includes at least the following operating condition data items: test type, capacitor model, test temperature, surge voltage and ripple current applied during the test, as well as observation data items such as test time and sample status; Based on the standardized sample data, the working condition data items are used as working condition keys for grouping, and the samples under each working condition are sorted according to the test time. The number of non-failed samples and the number of failed samples at each test time are counted in combination with the sample status, and a grouped censored sample containing interval censored samples and right censored samples is constructed. We construct feature vectors and use a Weibull regression model to characterize the capacitor lifetime distribution, where the scale parameter varies with operating conditions and the shape parameter is shared across different operating conditions. A negative log-likelihood objective function is established based on the grouped censored samples, wherein the probability contribution of the negative log-likelihood objective function is determined by the probability of the interval censored sample becoming invalid between adjacent detection times and the probability that the right censored sample is still valid at the cutoff observation time. The negative log-likelihood objective function is optimized and solved to estimate the parameters of the Weibull regression model, which include at least the shared shape parameter. Based on the trained Weibull regression model, the reliability curve and lifetime index under the target operating condition are calculated. When the target operating condition is determined to be a fully right-censored operating condition, the curve start-shift time is determined based on the cutoff observation time of the target operating condition. Before the curve start-shift time, a curve segment with a reliability of 1 is output, and after the curve start-shift time, a reliability curve segment calculated based on the Weibull regression model is output.

2. The method according to claim 1, characterized in that, The process of grouping, statistically analyzing, and sorting the standardized sample data according to the operating condition data items to construct a grouped censored sample that includes interval censored samples and right censored samples includes: The standardized sample data is grouped using the test type, capacitor model, test temperature, applied surge voltage, and applied ripple current as operating condition keys to obtain a sample set corresponding to each operating condition; the sample sets under each operating condition are then sorted according to the test time to obtain a time series corresponding to each operating condition. The samples under each working condition are sorted by time, and the time interval in which the failed sample falls is determined based on the change in the number of non-failed samples between adjacent detection times. Samples whose failure time falls between two adjacent detection times are recorded as interval censored samples. Samples for which no failure was observed up to the last detection time of a single operating condition are recorded as right-censored samples; When the total number of failures in a single operating condition is zero, the operating condition is marked as a fully right-censored operating condition.

3. The method according to claim 1, characterized in that, The construction of the feature vector includes: The feature vector is constructed by combining test type dummy variables, capacitor model dummy variables, and continuous stress characteristics; wherein, the continuous stress characteristics include at least the standardized test temperature characteristics, the surge voltage characteristics applied during the test, and the ripple current characteristics applied during the test; The Weibull regression model satisfies: in, The scale parameters corresponding to the target working condition. Shape parameters to be shared across different operating conditions; The continuous stress characteristics include at least temperature normalization characteristics. Surge voltage normalization characteristics and ripple current normalization characteristics ; The temperature normalization feature satisfies ; The surge voltage normalization characteristic satisfies ; The ripple current normalization characteristic satisfies ; in, For the test temperature, The surge voltage applied to the test, The ripple current applied to the test, , and These are the preset reference temperature, reference surge voltage, and reference ripple current, respectively.

4. The method according to any one of claims 1 to 3, characterized in that, The negative log-likelihood objective function is formed by accumulating the interval censored likelihood term and the right censored likelihood term for each working condition and time interval, and then adding a parameter regularization term. The interval censoring likelihood term is constructed based on the condition failure probability and the number of failure samples within the corresponding time interval. The condition failure probability is determined by the survival function values ​​of the Weibull regression model at the current time node and the previous time node. The right-censored likelihood term is constructed based on the survival probability at the cutoff time.

5. The method according to claim 1, characterized in that, The optimization of the negative log-likelihood objective function to estimate the parameters of the Weibull regression model includes: The negative log-likelihood objective function is iteratively optimized using multiple sets of different initial values ​​for the shared shape parameters. From the results of multiple optimizations, the set of parameters that minimizes the negative log-likelihood objective function is selected as the final estimated model parameters.

6. The method according to claim 1, characterized in that, The standardization process for the original data includes processing the identified missing data using different rules based on whether the missing field belongs to a preset key field. The key fields include the test type, capacitor model, test temperature, surge voltage applied during the test, and ripple current applied during the test for defining the operating condition key; the test time for time sorting and censoring interval division; and the sample status for sample health status identification. When the missing field belongs to a preset key field, the rules adopted include: removing the corresponding sample, retrospectively recording based on experimental data, or restoring the field value based on the fixed attributes of the same batch of samples; When the missing field does not belong to the preset key fields, the rules adopted include: using the mode of the same working condition to complete, the median of the same working condition to complete, the interpolation of the nearest time to complete, or the preset default value to complete, and generating the corresponding missing identification variable.

7. The method according to claim 1, characterized in that, When one or more preset complete working condition fields are missing from the simplified data file used for modeling, the standardization process for the original data also includes: The missing fields are recovered based on the external data source, and the recovered fields are used as part of the operating condition data items for subsequent processing; The external data source is at least one of the following: test data, sample master data table, product specification sheet, equipment configuration file, or fixed attributes of samples in the same batch.

8. The method according to claim 1, characterized in that, When the target operating condition is determined to be a fully right-censored operating condition, the reliability curve is output after a right-shift correction, including: When the target condition is marked as the fully right-censored condition, a curve start-up time greater than zero is determined based on the cutoff observation time of the condition and the preset correction rule. Before the curve shift time, output a curve segment with a reliability of 1; and after the curve shift time, use the difference between the current time and the curve shift time as the time input of the Weibull regression model, and output a right-shifted reliability curve segment. The life indicators under the target operating conditions include: B10 lifespan: ; Median lifespan: ; in, The dimensional parameters under the target working condition are... The shared shape parameters are used; when the right shift correction is not performed, the lifetime indicators include the B10 lifetime and median lifetime calculated based on the target operating condition scale parameters and the shared shape parameters; when the right shift correction is performed, the corrected B10 lifetime and the corrected median lifetime are obtained by adding the curve shift time and the corresponding Weibull lifetime quantile time, respectively.

9. The method according to claim 1, characterized in that, The standardization process for the original data further includes: Fields in the original data that have inconsistent names but equivalent meanings are uniformly mapped to standard field names through a preset field alias mapping table; Convert the category field in the original data into a categorical variable; The original test times in the raw data are uniformly converted into values ​​on a preset time scale; The specific operating condition data items include: The test type, capacitor model, test temperature, applied surge voltage, and applied ripple current are specified in the observation data items, which specifically include test time and sample status.

10. A capacitor accelerated aging reliability analysis system based on censored data processing, characterized in that, include: The data preprocessing module is used to acquire the raw data of the accelerated aging test of the capacitor and to standardize the raw data to obtain standardized sample data. The raw data includes at least the test type, capacitor model, test temperature, surge voltage applied during the test and ripple current applied during the test, as well as the test time and the observation data of the sample status. The grouped censoring construction module is used to group the standardized sample data with the working condition data item as the working condition key, sort the samples under each working condition according to the test time, and combine the sample status to count the number of non-failed samples and the number of failed samples at each detection time to construct a grouped censored sample that includes interval censored samples and right censored samples. The model building module is used to construct feature vectors and uses a Weibull regression model with scale parameters varying with operating conditions and shape parameters shared across different operating conditions to characterize the capacitor lifetime distribution. The parameter estimation module is used to establish a negative log-likelihood objective function based on the grouped censored samples, and to optimize and solve the negative log-likelihood objective function to estimate the parameters of the Weibull regression model, wherein the parameters include at least the shared shape parameter; wherein the probability contribution is determined based on the interval censoring probability of the interval censored samples and the survival probability of the right censored samples; The prediction and correction module is used to calculate the reliability curve and lifetime index under the target operating condition based on the trained Weibull regression model, and when the target operating condition is determined to be a fully right-censored operating condition, the reliability curve is corrected by right shift and then output.

Images