A method and system for multi-environment aging detection and evaluation of electrical devices

Abstract

This invention relates to the field of power electronics technology, and more particularly to a method and system for multi-environment aging detection and evaluation of electrical devices. The method includes the following steps: S1: Collecting environmental factor data and personnel activity data of gallium nitride (GaN) power devices in various application scenarios, and generating a test environment stress sequence based on the environmental factor data and personnel activity data; S2: Testing the GaN power devices according to the test environment stress sequence, and monitoring the aging parameters of the GaN power devices in real time; S3: Generating an aging feature vector based on the aging parameters, and determining the aging mechanism and confidence level based on the aging feature vector and an aging mechanism determination model. This invention determines the aging mechanism of devices under different application scenarios by simulating environmental factors and personnel activity data in different application scenarios, providing decision support for technicians engaged in device design and maintenance.

Description

Technical Field

[0001] This invention relates to the field of power electronics technology, and in particular to a method and system for multi-environment aging detection and evaluation of electrical devices. Background Technology

[0002] Against the backdrop of rapid development in electrical and electronic technologies, gallium nitride (GaN) power devices, as representatives of high-performance semiconductors, are widely used in electric vehicles, communication base stations, and industrial power supplies. However, in complex and ever-changing real-world application scenarios, these devices often experience accelerated aging due to the combined effects of environmental factors and human activities, leading to performance degradation issues such as threshold voltage drift and increased on-resistance. As GaN devices evolve towards higher power density and higher frequency, their aging process often involves multiple intertwined mechanisms. Traditional methods struggle to fully capture the dynamic stresses in real-world environments, resulting in biased lifetime predictions and increased risks of unexpected failures. Existing technologies mainly include accelerated aging tests based on single environmental factors, such as thermal aging experiments in constant temperature and humidity chambers or tracking device performance changes using electrical parameter monitoring equipment. While these methods can partially simulate the aging process, they are limited to fixed laboratory conditions and cannot effectively integrate the impact of random electrostatic discharge events generated by human activities. Therefore, there is an urgent need for a multi-environment aging detection and evaluation method and system for electrical devices that can simulate environmental factors and human activity data from different application scenarios for GaN power devices, determine the aging mechanisms of devices under different application scenarios, and provide decision support for technicians engaged in device design and maintenance. Summary of the Invention

[0003] To overcome the shortcomings of existing technologies in determining the aging mechanism of gallium nitride power devices in different application scenarios, this invention provides a multi-environment aging detection and evaluation method and system for electrical devices.

[0004] The technical implementation scheme of the present invention is: a multi-environment aging detection and evaluation method for electrical devices, comprising the following steps:

[0005] S1: Collect environmental factor data and personnel activity data in various application scenarios of gallium nitride power devices, and generate a test environment stress sequence based on the environmental factor data and personnel activity data;

[0006] S2: Test the gallium nitride power device according to the stress sequence of the test environment, and monitor the aging parameters of the gallium nitride power device in real time;

[0007] S3: Generate an aging feature vector based on the aging parameters, and determine the aging mechanism and confidence level based on the aging feature vector and the aging mechanism determination model;

[0008] S4: Record the stress sequence, aging feature vector, aging mechanism and confidence level of the test environment, and generate an aging test evaluation report.

[0009] Preferably, the step of collecting environmental factor data and personnel activity data in various application scenarios of gallium nitride power devices, and generating a test environment stress sequence based on the environmental factor data and personnel activity data, includes: collecting environmental factor data and personnel activity data of gallium nitride power devices in different application scenarios, wherein the environmental factor data includes scenario influencing factors, temperature, voltage, current, humidity, and vibration, and the personnel activity data includes personnel reference electrostatic voltage, comprehensive personnel activity intensity, activity frequency, and clothing material; analyzing the temporal changes of the environmental factor data to obtain a real environmental stress sequence; calculating the electrostatic discharge voltage based on the environmental factor data and the personnel activity data; generating a random electrostatic discharge event sequence through a Poisson distribution based on the electrostatic discharge voltage and the activity frequency; and spatiotemporally integrating the real environmental stress sequence and the electrostatic discharge sequence to generate a test environment stress sequence.

[0010] Preferably, the step of calculating the electrostatic discharge voltage based on the environmental factor data and the personnel activity data includes: calculating the electrostatic discharge voltage using an electrostatic discharge voltage calculation formula based on the scene influencing factors, humidity, personnel baseline electrostatic voltage, personnel comprehensive activity intensity, and clothing material, wherein the electrostatic discharge voltage calculation formula is:

[0011] ;

[0012] In the formula, This is the electrostatic discharge voltage. The reference electrostatic voltage for personnel, For the overall activity intensity of personnel, For clothing material coefficient, For humidity, This is the humidity adjustment constant. This refers to the scene's influencing factors.

[0013] Preferably, the step of testing the gallium nitride power device according to the test environment stress sequence and monitoring the aging parameters of the gallium nitride power device in real time includes: conducting a multi-factor aging experiment on the gallium nitride power device under the test environment stress sequence, and collecting the changes in the aging parameters of the device in real time to obtain an aging parameter change sequence, wherein the aging parameter change sequence includes a threshold voltage change sequence, a dynamic on-resistance change sequence, a static on-resistance change sequence, a thermal resistance change sequence, a leakage current change sequence, and a switching duration change sequence.

[0014] Preferably, the step of generating an aging feature vector based on the aging parameters and determining the aging mechanism and confidence level based on the aging feature vector and the aging mechanism determination model includes: performing time series smoothing preprocessing on the collected aging parameter change sequence, using wavelet transform to eliminate measurement errors, calculating the recoverable aging coefficient based on the dynamic on-resistance change sequence, and calculating the static on-resistance change rate, threshold voltage drift rate, leakage current change rate, thermal resistance change rate, and switching time delay change after normalizing the aging parameter change sequence to form an aging feature vector.

[0015] Preferably, the step of calculating the recoverable aging coefficient based on the dynamic on-resistance change sequence includes: dividing the dynamic on-resistance change sequence into an aging process sequence segment and a recovery process sequence segment using a moving window statistical method; obtaining the initial value of the dynamic on-resistance and the stable value of the dynamic on-resistance under stress from the aging process sequence segment; obtaining the stable value of the dynamic on-resistance after recovery within a preset recovery time window from the recovery process sequence segment; obtaining the recovery rate of the gallium nitride power device trap characteristics using a nonlinear least squares method to fit the recovery process sequence segment; and calculating the recoverable aging coefficient based on the recoverable aging coefficient calculation formula, wherein the recoverable aging coefficient calculation formula is:

[0016] ;

[0017] In the formula, To recover the aging coefficient, This represents the stable value of the dynamic on-resistance under stress. To restore the stable value of the dynamic on-resistance, This is the initial value of the dynamic on-resistance. For the recovery rate of the trap feature of gallium nitride power devices, It is a non-zero constant.

[0018] Preferably, the step of determining the aging mechanism and confidence level based on the aging feature vector and the aging mechanism determination model includes: inputting the aging feature vector into the aging mechanism determination model for analysis to obtain the aging mechanism determination result and confidence level; constructing the aging mechanism determination model based on the fusion of SVM probability likelihood and Bayesian methods; and training the aging mechanism determination model through an aging mechanism knowledge base, wherein the aging mechanism knowledge base presets multiple typical aging modes including gate oxide breakdown, interface trap accumulation, solder joint fatigue, hot carrier effect, and electrostatic damage and their corresponding aging feature vectors.

[0019] Preferably, the step of constructing an aging mechanism determination model based on the fusion of SVM probability likelihood and Bayesian methods, and training the aging mechanism determination model through an aging mechanism knowledge base, includes: extracting aging feature vectors for various typical aging modes preset in the knowledge base; training an SVM classification model using a Gaussian radial basis kernel function; the SVM classification model outputs the determination result and decision function value for each sample; mapping the decision function value output by the SVM classification model using a Sigmoid function to obtain the posterior probability of each mechanism; calculating the posterior probability of each mechanism based on Bayesian inference; selecting the mechanism with the highest posterior probability as the main aging mechanism of the current sample; calculating the normalized difference between the highest and second-highest posterior probability values; and combining the entropy values ​​of the probability distributions of all competing mechanisms to obtain the confidence level of the determination result through weighted summation.

[0020] Preferably, the step of recording the test environment stress sequence, aging feature vector, aging mechanism and confidence level, and generating an aging test evaluation report includes: uniformly storing and timestamping the collected test environment stress sequence, aging feature vector, aging mechanism determination result and confidence level; performing fusion analysis on multiple rounds of test results for the same application scenario; plotting change curves showing the evolution trend of the aging mechanism based on the aging mechanism determination result and confidence level; and generating aging test evaluation reports for different application scenarios.

[0021] A multi-environment aging test and evaluation system for electrical devices includes:

[0022] Data acquisition module: Collects environmental factor data and personnel activity data in various application scenarios of gallium nitride power devices, and generates a test environment stress sequence based on the environmental factor data and personnel activity data;

[0023] Real-time monitoring module: Tests gallium nitride power devices according to the stress sequence of the test environment and monitors the aging parameters of gallium nitride power devices in real time;

[0024] Mechanism determination module: Generates an aging feature vector based on the aging parameters, and determines the aging mechanism and confidence level based on the aging feature vector and the aging mechanism knowledge base of gallium nitride power devices;

[0025] Report generation module: Records the stress sequence, aging feature vector, aging mechanism and confidence level of the test environment, and generates an aging test evaluation report.

[0026] The beneficial effects of this invention are as follows:

[0027] This invention achieves high-fidelity simulation of real-world application scenarios, improving the accuracy and comprehensiveness of aging tests. By innovatively collecting environmental factor data and personnel activity data simultaneously, and generating random electrostatic discharge event sequences based on Poisson distribution, it can incorporate random stresses introduced by human operation that are difficult to reproduce in the laboratory into the test sequence. This allows for the construction of a test environment stress sequence that approximates the actual working environment of the device, making the aging test results more practically instructive.

[0028] This invention establishes a multi-dimensional real-time monitoring system for aging parameters, enabling in-depth insights into device performance degradation. By monitoring multiple key aging parameters in real time, including threshold voltage, dynamic / static on-resistance, thermal resistance, leakage current, and switching duration, the system can comprehensively and from multiple perspectives capture the performance evolution trajectory of devices under different stresses. In particular, by calculating the recoverable aging coefficient, it can effectively distinguish between recoverable degradation such as trap filling and irreversible permanent damage such as interface state generation, providing data support for accurately determining the aging mechanism.

[0029] This invention constructs a decision model that integrates SVM probability likelihood and Bayesian inference, and trains it using a knowledge base with multiple typical aging patterns. This enables intelligent analysis and pattern recognition of complex, high-dimensional aging feature vectors. The output not only determines the aging mechanism but also provides confidence levels based on posterior probabilities, helping engineers judge diagnostic results and effectively avoiding misjudgments. Attached Figure Description

[0030] Figure 1 This is a flowchart of a multi-environment aging test and evaluation method for electrical devices according to the present invention;

[0031] Figure 2 This is a structural diagram of a multi-environment aging detection and evaluation system for electrical devices according to the present invention. Detailed Implementation

[0032] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0033] Example 1: A multi-environment aging test and evaluation method for electrical devices, such as Figure 1 As shown, it includes the following steps:

[0034] S1: Collect environmental factor data and personnel activity data in various application scenarios of gallium nitride power devices, and generate a test environment stress sequence based on the environmental factor data and personnel activity data;

[0035] Environmental factor data and personnel activity data of gallium nitride power devices in different application scenarios are collected. The environmental factor data includes scenario influencing factors, temperature, voltage, current, humidity, and vibration. The personnel activity data includes personnel reference electrostatic voltage, comprehensive activity intensity, activity frequency, and clothing material. The temporal changes of the environmental factor data are analyzed to obtain the real environmental stress sequence. The electrostatic discharge voltage is calculated based on the environmental factor data and the personnel activity data. A random electrostatic discharge event sequence is generated by using a Poisson distribution based on the electrostatic discharge voltage and the activity frequency. The real environmental stress sequence and the electrostatic discharge sequence are spatiotemporally integrated to generate the test environmental stress sequence.

[0036] Calculating the electrostatic discharge voltage based on the environmental data and the personnel activity data includes: calculating the electrostatic discharge voltage using the electrostatic discharge voltage calculation formula based on the scene influencing factors, humidity, personnel baseline electrostatic voltage, personnel comprehensive activity intensity, and clothing material, wherein the electrostatic discharge voltage calculation formula is:

[0037] ;

[0038] In the formula, This is the electrostatic discharge voltage. The reference electrostatic voltage for personnel, For the overall activity intensity of personnel, For clothing material coefficient, For humidity, This is the humidity adjustment constant. This refers to the scene's influencing factors.

[0039] It should be further explained that this embodiment specifically illustrates a complete technical solution for constructing a test environment stress sequence for gallium nitride power devices in diverse application environments. In the initial stage, comprehensive scenario parameter acquisition is performed, collecting environmental factor data and personnel activity data under different application scenarios. Environmental factor data includes key physical quantities such as scenario influencing factors, temperature parameters, voltage parameters, current parameters, humidity parameters, and vibration parameters. Scenario influencing factors quantify the inherent electrostatic discharge capability of different application scenarios; for example, scenarios with antistatic flooring or metal mesh flooring are considered well-grounded scenarios. Other parameter data are continuously acquired through a distributed sensor network. Personnel activity data covers personnel reference electrostatic voltage, comprehensive activity intensity, activity frequency, and clothing material characteristics. The collection of personnel activity data is achieved through IoT visual recognition.

[0040] After completing the basic data collection, the process proceeds to the time-series characteristic analysis stage of environmental factor data. This process employs an adaptive time-series analysis algorithm to process continuous monitoring records of parameters such as temperature, voltage, current, humidity, and vibration. Kalman filtering is used to eliminate measurement noise, Fast Fourier Transform is used to identify periodic characteristics, and an autoregressive integral moving average model is used to predict parameter evolution trends. Following data preprocessing, the electrostatic discharge voltage is calculated. The personnel baseline electrostatic voltage is used as the calculation benchmark, measured under controlled conditions using a standard electrostatic potentiometer. The average contact time between personnel and equipment is obtained through target identification and detection as the comprehensive activity intensity of the personnel. The clothing material coefficient is determined based on the material-electrostatic correspondence established according to international standards. The environmental humidity parameter and humidity adjustment constant together constitute a nonlinear humidity correction term. The humidity adjustment constant is derived through regression analysis of extensive experimental data. The scene influence factor is set based on expert experience to assess the grounding of different application scenarios. After normalizing the above data to eliminate dimensional influences, the electrostatic discharge voltage calculation result is obtained using the electrostatic discharge voltage calculation formula.

[0041] After obtaining the electrostatic discharge voltage calculation results, they are combined with the activity frequency parameters from personnel activity data, and a Poisson stochastic process model is applied to generate a random electrostatic discharge event sequence. An adaptive parameter estimation method is used to determine the characteristic parameters of the Poisson distribution, and Monte Carlo simulation is used to generate an event sequence that conforms to actual statistical laws. The generated sequence contains randomly distributed event occurrence times and corresponding discharge voltage amplitudes. The time interval of each event follows an exponential distribution, and the voltage amplitude is dynamically adjusted according to the electrostatic discharge voltage calculation results, thereby accurately reproducing the spatiotemporal distribution characteristics and intensity features of electrostatic discharge events in actual working conditions. The real environmental stress sequence and the random electrostatic discharge sequence are spatiotemporally integrated. The two sequences are time-aligned and the sampling rate is unified, with steady-state environmental stress as the background and transient electrostatic discharge events as pulse interference superimposed. Through this processing method, the final test environment stress sequence retains the continuous variation characteristics of real environmental stress while incorporating the impact characteristics of random electrostatic discharge events.

[0042] S2: Test the gallium nitride power device according to the stress sequence of the test environment, and monitor the aging parameters of the gallium nitride power device in real time;

[0043] Under the stress sequence of the test environment, a multi-factor aging experiment was conducted on the gallium nitride power device. The changes in the aging parameters of the device were collected in real time to obtain the aging parameter change sequence. The aging parameter change sequence includes the threshold voltage change sequence, the dynamic on-resistance change sequence, the static on-resistance change sequence, the thermal resistance change sequence, the leakage current change sequence, and the switching time change sequence.

[0044] It should be further explained that this embodiment specifically illustrates the complete implementation process of testing gallium nitride power devices according to the stress sequence of the test environment and monitoring their aging parameters in real time. Under the continuous action of the stress sequence of the test environment, a multi-factor aging experiment is carried out on the gallium nitride power devices. This experimental process precisely controls the application sequence and intensity ratio of each stress factor in the stress sequence of the test environment. During the experiment, the aging parameter change data of the gallium nitride power devices are collected in real time through high-precision measurement. These aging parameters have clear physical definitions and measurement specifications: threshold voltage refers to the gate voltage required for the device to start conducting under a specific drain current condition; dynamic on-resistance characterizes the impedance characteristics of the device in the switching state; static on-resistance reflects the impedance performance of the device in steady-state conduction; thermal resistance is used to quantify the heat transfer performance between the internal junction temperature of the device and the external environment; leakage current reflects the small current leakage of the device in the off state; and switching duration includes two key indicators: turn-on delay time and turn-off delay time.

[0045] During real-time monitoring, each aging parameter is acquired using a specific measurement method. Threshold voltage monitoring involves accurately measuring the gate voltage drift while maintaining a constant drain current. Dynamic on-resistance measurement captures voltage and current waveforms and calculates impedance changes during device switching transients using a high-frequency sampling circuit. Static on-resistance testing eliminates the influence of contact resistance using a four-wire measurement method when the device is fully on. Thermal resistance monitoring involves simultaneously acquiring the device junction temperature and ambient temperature, and calculating the thermal resistance based on power dissipation. Leakage current measurement detects minute current changes using a high-impedance current probe when the device is off. Switching duration is acquired by capturing the timing difference between the gate drive signal and the drain voltage waveform using a digital storage oscilloscope.

[0046] All aging parameter monitoring data are continuously recorded with millisecond-level time resolution, forming a complete parameter evolution database. This comprehensive real-time monitoring scheme can accurately capture the performance degradation trajectory of gallium nitride power devices under combined stress, providing sufficient data support for subsequent aging mechanism analysis.

[0047] S3: Generate an aging feature vector based on the aging parameters, and determine the aging mechanism and confidence level based on the aging feature vector and the aging mechanism determination model;

[0048] The acquired aging parameter change sequence is preprocessed with time series smoothing, and wavelet transform is used to eliminate measurement errors. The recoverable aging coefficient is calculated based on the dynamic on-resistance change sequence. After normalizing the aging parameter change sequence, the static on-resistance change rate, threshold voltage drift rate, leakage current change rate, thermal resistance change rate, and switching time delay change are calculated to form an aging feature vector.

[0049] The recoverable aging coefficient is calculated based on the dynamic on-resistance change sequence, including: dividing the dynamic on-resistance change sequence into an aging process sequence segment and a recovery process sequence segment using a moving window statistical method; obtaining the initial value of the dynamic on-resistance and the stable value of the dynamic on-resistance under stress from the aging process sequence segment; obtaining the stable value of the dynamic on-resistance after recovery within a preset recovery time window from the recovery process sequence segment; obtaining the recovery rate of the gallium nitride power device trap characteristics using a nonlinear least squares method to fit the recovery process sequence segment; and calculating the recoverable aging coefficient based on the recoverable aging coefficient calculation formula, wherein the recoverable aging coefficient calculation formula is:

[0050] ;

[0051] In the formula, To recover the aging coefficient, This represents the stable value of the dynamic on-resistance under stress. To restore the stable value of the dynamic on-resistance, This is the initial value of the dynamic on-resistance. For the recovery rate of the trap feature of gallium nitride power devices, It is a non-zero constant.

[0052] The aging feature vector is input into the aging mechanism determination model for analysis to obtain the aging mechanism determination result and confidence level. The aging mechanism determination model is constructed based on the fusion of SVM probability likelihood and Bayesian. The aging mechanism determination model is trained through an aging mechanism knowledge base. The aging mechanism knowledge base presets a variety of typical aging modes, including gate oxide layer breakdown, interface trap accumulation, solder joint fatigue, hot carrier effect and electrostatic damage and corresponding aging feature vectors.

[0053] For multiple typical aging modes preset in the knowledge base, aging feature vectors are extracted for each mode. An SVM classification model is trained using a Gaussian radial basis function kernel. The SVM classification model outputs the judgment result and decision function value for each sample. The decision function value output by the SVM classification model is probabilistically mapped using a Sigmoid function to obtain the posterior probability of each mechanism. The posterior probability of each mechanism is calculated based on Bayesian inference. The mechanism with the highest posterior probability is selected as the main aging mechanism of the current sample. The normalized difference between the highest and second-highest posterior probability values ​​is calculated. Combined with the entropy values ​​of the probability distributions of all competing mechanisms, a weighted sum is obtained to obtain the confidence level of the judgment result.

[0054] It should be further explained that this embodiment details the specific implementation of generating aging feature vectors based on aging parameters and using an aging mechanism determination model to determine the aging mechanism and confidence level. After obtaining the real-time monitored aging parameter change sequence, in the initial data processing stage, the acquired aging parameter change sequence needs to undergo time series smoothing preprocessing. This processing uses a moving average filtering technique, which establishes a sliding window of length N and performs convolution operations on the original observation data to effectively filter out random fluctuations and transient interference. The determination of the window size needs to comprehensively consider the relationship between the signal characteristic frequency and the sampling rate. Usually, a window length covering 3-5 characteristic periods is selected to achieve the optimal balance between smoothing effect and feature fidelity. Subsequently, wavelet transform is used to eliminate measurement errors.

[0055] In the calculation of the recoverable aging coefficient, a moving window statistical method is used to divide the dynamic on-resistance change sequence into stages. This method sets two independent time windows: the aging monitoring window is fixed at 120% of the stress duration, while the recovery monitoring window is determined based on the device's thermal time constant. By calculating the standard deviation of the data within the window, the critical point between the stress application stage and the recovery stage is accurately identified. When extracting the initial value of the dynamic on-resistance from the aging process sequence, the average value of the first 10 sampling periods before stress application is taken as the baseline value; the stable value of the dynamic on-resistance under stress is taken as the average value within ±2% of the fluctuation range at the end of the stress period. In the recovery process sequence, the preset recovery time window is determined based on the typical operating interval of the application scenario, and the average value of the last 5 sampling points within this time window is taken as the stable value of the dynamic on-resistance after recovery.

[0056] The recovery rate of gallium nitride power devices' trap characteristics is obtained by fitting a sequence of recovery process segments using a nonlinear least squares method. First, a physical model describing the recovery process is established, typically in the form of an exponential decay function. The model includes three parameters to be determined: the stable value of the dynamic on-resistance after recovery, the stable value of the dynamic on-resistance under stress, and the core trap characteristic recovery rate. The trap characteristic recovery rate describes the rate characteristics of trap trapping and releasing carriers; a small trap characteristic recovery rate indicates a slow recovery process, while a large value indicates a fast recovery process. The fitting process uses an iterative optimization algorithm to find the optimal parameter combination that minimizes the sum of squared residuals between the calculated model values ​​and the experimental measurements.

[0057] The implementation of the nonlinear least squares method involves several steps. First, initial estimates of the parameters are set, determined initially through piecewise linear regression. Then, an iterative calculation phase begins, where the model's predicted values ​​for the current parameters are calculated in each iteration and compared with actual observations to obtain the residuals. By calculating the sum of squares of the residuals and their partial derivatives with respect to each parameter, a gradient vector and a Hessian matrix are constructed, thus determining the adjustment direction and step size of the parameters. This process is repeated until a convergence condition is met, typically set to the condition that the change in the sum of squares of the residuals between two consecutive iterations is less than a preset threshold. This preset threshold is obtained through the experience and practice of experts in the field. The recoverable aging coefficient is calculated by normalizing the stable value of the dynamic on-resistance under stress, the stable value of the dynamic on-resistance after recovery, the initial value of the dynamic on-resistance, and the recovery rate of the gallium nitride power device trap characteristics.

[0058] Next, the variation sequences of each aging parameter are normalized, and based on this, the rate of change of static on-resistance, threshold voltage drift rate, leakage current rate of change, thermal resistance rate of change, and switching time delay are calculated. These standardized parameter variations collectively constitute a multi-dimensional aging feature vector characterizing the device's degradation state, where each dimension reflects the degree of aging of a specific performance parameter of the device. In the aging mechanism determination stage, the generated aging feature vector is input into a specially trained aging mechanism determination model for analysis, thereby obtaining accurate aging mechanism determination results and corresponding confidence levels. This aging mechanism determination model is built based on a fusion architecture of support vector machine probability likelihood and Bayesian inference, and is trained using a pre-set aging mechanism knowledge base. The aging mechanism knowledge base pre-establishes various typical aging modes and their corresponding aging feature vector mapping relationships. These typical aging modes include gate oxide breakdown, interface trap accumulation, solder joint fatigue, hot carrier effects, and common failure mechanisms such as electrostatic damage. In the model construction and training stage, the various typical aging modes pre-set in the knowledge base are processed. For typical aging modes such as gate oxide breakdown, interface trap accumulation, solder joint fatigue, hot carrier effect, and electrostatic damage, corresponding aging feature vectors were extracted from historical experimental data to serve as training sample sets. These training samples cover device degradation data under various stress conditions, ensuring the representativeness and completeness of the sample sets. In the classification model training stage, a Gaussian radial basis kernel function was used to construct a support vector machine (SVM) classification model. This kernel function transforms the original feature space to a high-dimensional space through nonlinear mapping, effectively handling the complex boundary problems between features of various aging modes. During model training, a sequential minimum optimization algorithm was used to solve the convex quadratic programming problem to obtain the optimal classification hyperplane. The trained SVM classification model outputs a preliminary class determination result and a corresponding decision function value for each input sample. These values ​​reflect the relative position of the sample to the classification hyperplane. To achieve probabilistic output, the decision function values ​​output by the SVM were nonlinearly transformed using the Sigmoid function. This transformation process establishes a mapping relationship from the decision function value to the posterior probability, enabling each aging mechanism to obtain a corresponding likelihood probability estimate. In the probability fusion stage, the likelihood probabilities of each aging mechanism are further optimized based on a Bayesian inference framework. First, prior probabilities for each aging mechanism are set based on historical data statistics. Then, combined with the likelihood probabilities output by the support vector machine, the posterior probability of each aging mechanism is calculated using Bayes' theorem. This process fully considers the combined influence of prior knowledge and current observation data, making the probability estimation more objective and accurate. In the final decision-making stage, by comparing the posterior probability values ​​of all candidate aging mechanisms, the aging mechanism with the highest posterior probability value is selected as the primary aging mechanism for the current test sample.Meanwhile, in order to quantify the confidence level of the judgment result, the normalized difference between the maximum posterior probability value and the second largest probability value is calculated, and the entropy values ​​of the probability distributions of all competition mechanisms are combined to calculate the confidence level of the judgment result by weighted summation.

[0059] S4: Record the stress sequence, aging feature vector, aging mechanism and confidence level of the test environment, and generate an aging test evaluation report.

[0060] The collected test environment stress sequence, aging feature vector, aging mechanism judgment results and confidence level are uniformly stored and timestamped. The test results of multiple rounds in the same application scenario are fused and analyzed. Based on the aging mechanism judgment results and confidence level, the change curves showing the evolution trend of the aging mechanism are plotted, and aging test evaluation reports for different application scenarios are generated.

[0061] It should be further explained that this embodiment fully describes the specific implementation process of test data recording and evaluation report generation. After the aging mechanism determination is completed, the collected test environment stress sequence, aging feature vector, aging mechanism determination results, and corresponding confidence levels are stored and managed. All data adopts a unified distributed storage architecture, and each data is attached with a precise timestamp. The timestamp recording format is accurate to the millisecond level to ensure strict correspondence and complete traceability of data time sequence. For multiple rounds of test results under the same application scenario, a data fusion analysis mechanism is initiated. This mechanism first aligns the time sequence data of each test, eliminates the data discontinuity caused by the measurement interval through a feature matching algorithm, and then integrates the aging feature vectors of multiple rounds of tests using a weighted fusion strategy, with the confidence index serving as an important basis for weight allocation. This longitudinal comparative analysis method can effectively identify the gradual law and abrupt characteristics of device performance degradation. Based on the dataset after fusion analysis, change curves showing the evolution trend of aging mechanism are automatically plotted. These curves are based on the time axis and visualize the determination results of different aging mechanisms and their confidence levels in the form of multi-dimensional waveforms. The curve generation process employs the least squares method for trend fitting while preserving the discrete distribution characteristics of the original data points, thus clearly showcasing short-term fluctuations and long-term trends. Specifically, for data segments with low confidence, the curve is marked with special symbols to indicate uncertainty in the judgment results for that period. Based on the complete analysis results, aging detection and evaluation reports are generated for different application scenarios. The reports use a structured presentation, detailing the specific measurement environment stress sequence, the evolution of aging characteristic parameters, the transformation path of the dominant aging mechanism, and the confidence assessment of the aging mechanism at key time points for each application scenario.

[0062] Example 2: Based on Example 1, a multi-environment aging detection and evaluation system for electrical devices, such as... Figure 2 As shown, it includes:

[0063] Data acquisition module: Collects environmental factor data and personnel activity data in various application scenarios of gallium nitride power devices, and generates a test environment stress sequence based on the environmental factor data and personnel activity data;

[0064] Real-time monitoring module: Tests gallium nitride power devices according to the stress sequence of the test environment and monitors the aging parameters of gallium nitride power devices in real time;

[0065] Mechanism determination module: Generates an aging feature vector based on the aging parameters, and determines the aging mechanism and confidence level based on the aging feature vector and the aging mechanism knowledge base of gallium nitride power devices;

[0066] Report generation module: Records the stress sequence, aging feature vector, aging mechanism and confidence level of the test environment, and generates an aging test evaluation report.

[0067] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for multi-environment aging detection and evaluation of electrical devices, characterized in that, Includes the following steps: S1: Collect environmental factor data and personnel activity data in various application scenarios of gallium nitride power devices, and generate a test environment stress sequence based on the environmental factor data and personnel activity data. This includes: collecting environmental factor data and personnel activity data of gallium nitride power devices in different application scenarios, wherein the environmental factor data includes scenario influencing factors, temperature, voltage, current, humidity, and vibration, and the personnel activity data includes personnel reference electrostatic voltage, comprehensive personnel activity intensity, activity frequency, and clothing material; analyzing the temporal changes of the environmental factor data to obtain a real environmental stress sequence; calculating the electrostatic discharge voltage based on the environmental factor data and the personnel activity data; generating a random electrostatic discharge event sequence through Poisson distribution based on the electrostatic discharge voltage and the activity frequency; and spatiotemporally integrating the real environmental stress sequence and the random electrostatic discharge event sequence to generate a test environment stress sequence. S2: Test the gallium nitride power device according to the stress sequence of the test environment, and monitor the aging parameters of the gallium nitride power device in real time; S3: Generate an aging feature vector based on the aging parameters, and determine the aging mechanism and confidence level based on the aging feature vector and the aging mechanism determination model; S4: Record the stress sequence, aging feature vector, aging mechanism and confidence level of the test environment, and generate an aging test evaluation report.

2. The multi-environment aging test and evaluation method for electrical devices according to claim 1, characterized in that, The calculation of electrostatic discharge voltage based on the environmental factor data and the personnel activity data includes: calculating the electrostatic discharge voltage using the electrostatic discharge voltage calculation formula based on the scene influencing factors, humidity, personnel baseline electrostatic voltage, personnel comprehensive activity intensity, and clothing material, wherein the electrostatic discharge voltage calculation formula is: ； In the formula, This is the electrostatic discharge voltage. The reference electrostatic voltage for personnel, For the overall activity intensity of personnel, For clothing material coefficient, For humidity, This is the humidity adjustment constant. This refers to the scene's influencing factors.

3. The multi-environment aging test and evaluation method for electrical devices according to claim 1, characterized in that, The step of testing gallium nitride power devices according to the test environment stress sequence and monitoring the aging parameters of gallium nitride power devices in real time includes: conducting multi-factor aging experiments on gallium nitride power devices under the test environment stress sequence, and collecting the changes in aging parameters of the devices in real time to obtain an aging parameter change sequence. The aging parameter change sequence includes a threshold voltage change sequence, a dynamic on-resistance change sequence, a static on-resistance change sequence, a thermal resistance change sequence, a leakage current change sequence, and a switching time change sequence.

4. The multi-environment aging test and evaluation method for electrical devices according to claim 3, characterized in that, The step of generating an aging feature vector based on the aging parameters and determining the aging mechanism and confidence level based on the aging feature vector and the aging mechanism determination model includes: performing time series smoothing preprocessing on the collected aging parameter change sequence, using wavelet transform to eliminate measurement errors, calculating the recoverable aging coefficient based on the dynamic on-resistance change sequence, and calculating the static on-resistance change rate, threshold voltage drift rate, leakage current change rate, thermal resistance change rate, and switching time delay change after normalizing the aging parameter change sequence to form the aging feature vector.

5. The multi-environment aging test and evaluation method for electrical devices according to claim 4, characterized in that, The step of calculating the recoverable aging coefficient based on the dynamic on-resistance change sequence includes: dividing the dynamic on-resistance change sequence into an aging process sequence segment and a recovery process sequence segment using a moving window statistical method; obtaining the initial value of the dynamic on-resistance and the stable value of the dynamic on-resistance under stress from the aging process sequence segment; obtaining the stable value of the dynamic on-resistance after recovery within a preset recovery time window from the recovery process sequence segment; obtaining the recovery rate of the gallium nitride power device trap characteristics using a nonlinear least squares method based on the recovery process sequence segment; and calculating the recoverable aging coefficient based on the recoverable aging coefficient calculation formula, wherein the recoverable aging coefficient calculation formula is: ； In the formula, To recover the aging coefficient, This represents the stable value of the dynamic on-resistance under stress. To restore the stable value of the dynamic on-resistance, This is the initial value of the dynamic on-resistance. For the recovery rate of the trap feature of gallium nitride power devices, It is a non-zero constant.

6. The multi-environment aging test and evaluation method for electrical devices according to claim 1, characterized in that, The step of determining the aging mechanism and confidence level based on the aging feature vector and the aging mechanism determination model includes: inputting the aging feature vector into the aging mechanism determination model for analysis to obtain the aging mechanism determination result and confidence level; constructing the aging mechanism determination model based on the fusion of SVM probability likelihood and Bayesian methods; and training the aging mechanism determination model through an aging mechanism knowledge base. The aging mechanism knowledge base pre-sets multiple typical aging modes, including gate oxide breakdown, interface trap accumulation, solder joint fatigue, hot carrier effect, and electrostatic damage, and their corresponding aging feature vectors.

7. The multi-environment aging test and evaluation method for electrical devices according to claim 6, characterized in that, The aging mechanism determination model, constructed by fusing SVM probability likelihood and Bayesian methods, is trained using an aging mechanism knowledge base. This process includes: extracting aging feature vectors for various typical aging modes preset in the knowledge base; training an SVM classification model using a Gaussian radial basis function kernel; outputting the determination result and decision function value for each sample using the SVM classification model; mapping the decision function value output by the SVM classification model using a Sigmoid function to obtain the posterior probability of each mechanism; calculating the posterior probability of each mechanism based on Bayesian inference; selecting the mechanism with the highest posterior probability as the primary aging mechanism for the current sample; calculating the normalized difference between the highest and second-highest posterior probability values; and combining the entropy values ​​of the probability distributions of all competing mechanisms to obtain the confidence level of the determination result through weighted summation.

8. The multi-environment aging test and evaluation method for electrical devices according to claim 1, characterized in that, The process of recording the stress sequence, aging feature vector, aging mechanism, and confidence level of the test environment and generating an aging test evaluation report includes: uniformly storing and timestamping the collected stress sequence, aging feature vector, aging mechanism determination results, and confidence level of the test environment; performing fusion analysis on multiple rounds of test results for the same application scenario; plotting change curves showing the evolution trend of the aging mechanism based on the aging mechanism determination results and confidence level; and generating aging test evaluation reports for different application scenarios.

9. A multi-environment aging test and evaluation system for electrical devices, used to implement the multi-environment aging test and evaluation method for electrical devices as described in any one of claims 1-8, characterized in that, include: Data acquisition module: Collects environmental factor data and personnel activity data in various application scenarios of gallium nitride power devices, and generates a test environment stress sequence based on the environmental factor data and personnel activity data; Real-time monitoring module: Tests gallium nitride power devices according to the stress sequence of the test environment and monitors the aging parameters of gallium nitride power devices in real time; Mechanism determination module: Generates an aging feature vector based on the aging parameters, and determines the aging mechanism and confidence level based on the aging feature vector and the aging mechanism knowledge base of gallium nitride power devices; Report generation module: Records the stress sequence, aging feature vector, aging mechanism and confidence level of the test environment, and generates an aging test evaluation report.

Images