Method for testing aging degree of high-frequency switching power supply

Abstract

The invention discloses a method for testing the aging degree of a high-frequency switching power supply, relates to the technical field of power supply aging degree testing, and realizes device-level aging identification and residual life prediction of a capacitor, a transformer and a power device of the high-frequency switching power supply through online micro-disturbance excitation, multi-source dynamic response acquisition and digital twin parameter inversion. And generating a rack-level aging risk distribution and load scheduling strategy, and completing refined aging evaluation and planned maintenance decision under the condition of no shutdown and no disassembly.

Description

Technical Field

[0001] This invention relates to the field of power supply aging test technology, and in particular to a method for testing the aging of high-frequency switching power supplies. Background Technology

[0002] High-frequency switching power supplies are widely used in 5G communication equipment rooms, data center rack power supplies, and operator core network power supply systems. They typically operate in parallel groups of 48V high-frequency switching rectifier modules, enduring long-term high load, high temperature fluctuations, and frequent start-stop conditions. Accumulated issues such as electrolytic capacitor drying out, power device junction temperature cycling fatigue, transformer insulation aging, and control loop parameter drift eventually manifest as poor transient response, increased ripple and electromagnetic radiation, and abnormal temperature rise. However, under rated steady-state output, they often maintain voltage and current within allowable ranges, making them difficult to detect using traditional monitoring methods.

[0003] In existing technologies, operation and maintenance personnel mainly assess the status of high-frequency switching power supplies through two methods: one is to monitor a small number of steady-state indicators such as output voltage, current, and alarm status in the data center monitoring system, and make a rough judgment based on whether they exceed the limits; the other is to remove a small number of power supplies for testing during planned power outages or annual maintenance, and conduct sampling assessments using offline withstand voltage tests, capacitor ESR measurements, and temperature rise tests. These methods have obvious limitations: they cannot identify modules that are "about to fail" without shutting down the system, they cannot distinguish between different failure mechanisms mainly caused by capacitor aging, transformer aging, or power device fatigue, and they are also difficult to provide quantitative assessment results of remaining life.

[0004] Currently, Chinese patent application number 202410202979.4 discloses a testing method, apparatus, computer equipment, and storage medium for portable power banks. The method includes: discharging a power bank under test using a test power bank until it is fully charged; determining the remaining capacity of the power bank under test based on the discharge amount of the test power bank; determining the capacity reduction of the power bank under test based on the remaining capacity and the rated capacity of the test power bank; charging both the test power bank and the power bank under test separately using the same power supply; determining a first charging speed of the test power bank and a second charging speed of the power bank under test; determining the charging speed reduction of the power bank under test based on the first and second charging speeds; and determining the aging index of the power bank under test based on the capacity reduction and the charging speed reduction, wherein the aging index is directly proportional to the degree of aging of the power bank under test. This application effectively achieves the aging degree testing of portable power banks.

[0005] The relevant technologies are insufficient to achieve precise prediction of device-level aging and remaining lifespan of high-frequency switching power supplies based on multi-source dynamic response without shutting down or disassembling the power supply. Summary of the Invention

[0006] The technical problem solved by this invention is that existing technologies are unable to achieve precise prediction of device-level aging identification and remaining lifespan of high-frequency switching power supplies based on multi-source dynamic response without shutting down or disassembling the power supply.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution: A method for testing the aging degree of a high-frequency switching power supply includes the following steps: Step S1: Retrieve high-frequency switching power supply aging sample data from the aging sample database, establish a digital twin baseline model containing aging-sensitive parameters, and output the first analysis result; Step S2: Collect online steady-state operation data of the high-frequency switching power supply under test in the target application scenario and perform preprocessing to form an online steady-state operation dataset; Step S3: Apply micro-perturbation excitation to the high-frequency switching power supply under test within the preset test time window, collect dynamic response operation data and form an online dynamic response dataset; Step S4: Input the online steady-state operation dataset and the online dynamic response dataset into the digital twin baseline model, obtain the aging-sensitive parameters through parameter inversion and generate the aging characteristic index set, and output the second analysis result; Step S5: Evaluate the aging degree and remaining lifespan of the high-frequency switching power supply under test based on the first and second analysis results, generate an aging evaluation result set and load scheduling strategy, and input them to the control terminal.

[0008] Preferably, step S1 includes the following sub-steps: Step S101: Retrieve high-frequency switching power supply aging sample data from the aging sample database. The aging sample data includes output voltage time series, output current time series, ripple noise time series, electromagnetic radiation spectrum data and key component temperature time series under multiple aging stages, as well as corresponding aging stage labels, to form an aging sample dataset. Step S102: Establish a parameterized digital twin baseline model based on the high-frequency switching power supply circuit topology. The parameters of the digital twin baseline model include the equivalent series resistance of the electrolytic capacitor, the equivalent inductance value, the leakage inductance of the transformer, the on-resistance of the power device, and the control loop gain, forming a baseline model parameter set. Step S103: Input the aging sample dataset into the digital twin baseline model, and calculate the aging sensitivity parameter values ​​corresponding to each aging stage by fitting the consistency between the simulation waveform and simulation spectrum under each aging stage and the aging sample dataset. Establish the mapping relationship between the aging sensitivity parameters and the aging stage, and output the first analysis result.

[0009] Preferably, step S2 includes the following sub-steps: Step S201: Connect the high-frequency switching power supply under test to the load system of the target application scenario. Through voltage and current sampling circuit, ripple acquisition circuit, electromagnetic radiation probe and temperature sensor, online acquisition of output voltage time series, output current time series, ripple noise time series, electromagnetic radiation spectrum data and key component temperature time series are formed to form online raw steady state dataset. Step S202: Time base alignment, outlier removal, and noise filtering are performed on the online raw steady-state dataset. The aligned output voltage time series, output current time series, ripple noise time series, electromagnetic radiation spectrum data, and key device temperature time series are combined to form an online steady-state operation dataset.

[0010] Preferably, step S3 includes the following sub-steps: Step S301: Within the preset test time window, apply a small step disturbance and a frequency sweep sinusoidal disturbance to the load current near the rated load of the high-frequency switching power supply under test through the load modulation unit, record the time series of the disturbance signal, and form a micro-disturbance excitation dataset. Step S302: Under the action of the micro-perturbation excitation dataset, the output voltage response waveform, output current response waveform, ripple noise change sequence, electromagnetic radiation spectrum change data and key component temperature change time series of the high-frequency switching power supply under test are collected to form an online raw dynamic dataset. Step S303: Time alignment and feature extraction are performed on the online original dynamic dataset. The overshoot, settling time, and oscillation decay characteristics of the output voltage and output current are calculated. The ripple spectrum energy distribution index, electromagnetic radiation spectrum amplitude index, and temperature rise rate index are calculated and combined to form an online dynamic response dataset.

[0011] Preferably, step S4 includes the following sub-steps: Step S401: Input the online steady-state operation dataset and the online dynamic response dataset into the digital twin baseline model, and use the parameter inversion algorithm to iteratively adjust the equivalent series resistance of the electrolytic capacitor, the equivalent inductance value, the leakage inductance of the transformer, the on-resistance of the power device, and the control loop gain in the parameter set of the baseline model; Step S402: In each iteration, calculate the fitting error between the output voltage time series, output current time series, ripple spectrum, electromagnetic radiation spectrum, and key device temperature change and the online steady-state operation dataset and online dynamic response dataset obtained from the digital twin baseline model simulation. Stop the iteration when the fitting error is lower than the preset threshold to obtain the inverted aging sensitive parameter set. Step S403: Based on the offset ratio between the inverted aging sensitive parameter set and the baseline model parameter set, calculate the capacitor aging index, transformer aging index, power device fatigue index, and control loop degradation index, combine the indices to form an aging characteristic index set, and output it as the second analysis result.

[0012] Preferably, step S5 includes the following sub-steps: Step S501: Input the aging characteristic index set in the second analysis result into the lifetime prediction model trained based on the aging sample dataset. Calculate the remaining lifetime estimate of the high-frequency switching power supply under test based on the mapping relationship between capacitor aging index, transformer aging index, power device fatigue index, control loop degradation index and aging stage. Step S502: Based on the aging characteristic index set and the remaining life estimate, the high-frequency switching power supply under test is divided into multiple aging levels according to the preset aging level threshold, and an aging assessment result set is generated. The aging assessment result set includes the aging level, aging percentage and remaining life estimate. Step S503: In rack application scenarios, the aging assessment result sets of multiple high-frequency switching power supplies under test are compared horizontally to generate rack-level aging risk distribution data. Based on the rack-level aging risk distribution data, the load scheduling strategy and replacement recommendations are determined and input to the control terminal.

[0013] Preferably, the logic for obtaining the micro-perturbation excitation dataset is as follows: The amplitude of load step disturbance near the rated load is limited to a preset percentage of the rated load; The frequency range of the sweeping sinusoidal disturbance is set to cover the bandwidth of the high-frequency switching power supply control loop and its nearby frequency bands, and the preset duration is maintained at each frequency point. The amplitude and frequency distribution of the micro-perturbation excitation dataset are adaptively adjusted based on the observability indices of the output voltage and output current responses.

[0014] Preferably, the parameter inversion algorithm includes: A weighted objective function is constructed with the fitting error of the output voltage time series, the fitting error of the output current time series, the fitting error of the ripple spectrum, the fitting error of the electromagnetic radiation spectrum, and the fitting error of the temperature change of key components as components; In each iteration, the equivalent series resistance of the electrolytic capacitor, the equivalent inductance value, the leakage inductance of the transformer, the on-resistance of the power device, and the gain of the control loop are adjusted simultaneously according to the gradient of the objective function or the search direction. When the weighted objective function is less than the preset convergence threshold or the number of iterations reaches the upper limit, the set of inverted aging sensitive parameters is output.

[0015] Preferably, the logic for determining the degree of aging is as follows: The aging level is preset, and each level corresponds to the threshold range of capacitor aging index, transformer aging index, power device fatigue index and control loop degradation index. Based on the position of each index in the aging characteristic index set within the corresponding threshold range, a weighted comprehensive score is calculated, and the comprehensive score is mapped to the corresponding aging level. When any single index exceeds its highest level threshold, the high-frequency switching power supply under test will be directly determined to be at the highest aging level and marked as a high-risk object in the aging assessment results set.

[0016] Preferably, the logic for generating the load scheduling strategy is as follows: Based on rack-level aging risk distribution data, health priorities are assigned to each high-frequency switching power supply. Power supplies with low aging levels and high remaining life estimates are classified as high priority, while power supplies with high aging levels or low remaining life estimates are classified as low priority. Under the constraint of meeting the total rack power requirements, more load is allocated to high-priority power supplies, and the load of low-priority power supplies is limited to below the preset upper limit. When the estimated remaining life of a certain high-frequency switching power supply in the aging assessment results is lower than the preset maintenance time window, it is marked as a replacement object in the load scheduling strategy, and a corresponding planned replacement suggestion is generated.

[0017] The beneficial effects of this invention are as follows: This invention achieves device-level aging identification and remaining lifetime prediction of capacitors, transformers and power devices in high-frequency switching power supplies through online micro-perturbation excitation, multi-source dynamic response acquisition and digital twin parameter inversion, generates rack-level aging risk distribution and load scheduling strategy, and completes refined aging assessment and planned maintenance decision-making without stopping the machine or disassembling the machine. Attached Figure Description

[0018] Figure 1 The flowchart illustrates the steps of a high-frequency switching power supply aging test method according to an embodiment of the present invention. Detailed Implementation

[0019] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0020] Example, refer to Figure 1 A method for testing the aging degree of a high-frequency switching power supply is provided, comprising the following steps: Step S1: Retrieve high-frequency switching power supply aging sample data from the aging sample database, establish a digital twin baseline model containing aging-sensitive parameters, and output the first analysis result.

[0021] Step S2: Collect online steady-state operation data of the high-frequency switching power supply under test in the target application scenario and perform preprocessing to form an online steady-state operation dataset.

[0022] Step S3: Apply micro-perturbation excitation to the high-frequency switching power supply under test within the preset test time window, collect dynamic response operation data, and form an online dynamic response dataset.

[0023] Step S4: Input the online steady-state operation dataset and the online dynamic response dataset into the digital twin baseline model, obtain the aging-sensitive parameters through parameter inversion and generate the aging characteristic index set, and output the second analysis result.

[0024] Step S5: Evaluate the aging degree and remaining lifespan of the high-frequency switching power supply under test based on the first and second analysis results, generate an aging evaluation result set and load scheduling strategy, and input them to the control terminal.

[0025] This invention achieves device-level aging identification and remaining lifetime prediction for capacitors, transformers, and power devices in high-frequency switching power supplies through online micro-perturbation excitation, multi-source dynamic response acquisition, and digital twin parameter inversion. It generates rack-level aging risk distribution and load scheduling strategies, enabling refined aging assessment and planned maintenance decisions without shutting down or disassembling the equipment.

[0026] Step S1 includes the following sub-steps: Step S101: Retrieve high-frequency switching power supply aging sample data from the aging sample database. The aging sample data includes output voltage time series, output current time series, ripple noise time series, electromagnetic radiation spectrum data and key component temperature time series under multiple aging stages, as well as corresponding aging stage labels, to form an aging sample dataset.

[0027] Step S102: Establish a parameterized digital twin baseline model based on the high-frequency switching power supply circuit topology. The parameters of the digital twin baseline model include the equivalent series resistance of the electrolytic capacitor, the equivalent inductance value, the leakage inductance of the transformer, the on-resistance of the power device, and the control loop gain, forming a baseline model parameter set.

[0028] Step S103: Input the aging sample dataset into the digital twin baseline model, and calculate the aging sensitivity parameter values ​​corresponding to each aging stage by fitting the consistency between the simulation waveform and simulation spectrum under each aging stage and the aging sample dataset. Establish the mapping relationship between the aging sensitivity parameters and the aging stage, and output the first analysis result.

[0029] Step S2 includes the following sub-steps: Step S201: Connect the high-frequency switching power supply under test to the load system of the target application scenario. Through voltage and current sampling circuit, ripple acquisition circuit, electromagnetic radiation probe and temperature sensor, online acquisition of output voltage time series, output current time series, ripple noise time series, electromagnetic radiation spectrum data and key component temperature time series are formed to form an online raw steady-state dataset.

[0030] Step S202: Time base alignment, outlier removal, and noise filtering are performed on the online raw steady-state dataset. The aligned output voltage time series, output current time series, ripple noise time series, electromagnetic radiation spectrum data, and key device temperature time series are combined to form an online steady-state operation dataset.

[0031] Step S3 includes the following sub-steps: Step S301: Within a preset test time window, a small step disturbance and a frequency sweep sinusoidal disturbance are applied to the load current near the rated load of the high-frequency switching power supply under test by the load modulation unit, and the time series of the disturbance signal is recorded to form a micro-disturbance excitation dataset.

[0032] The rated output current of the high-frequency switching power supply under test is denoted as... The current steady-state load current is denoted as .

[0033] At the preset time When a small step load is applied, the load current after the step is... for: (when ); (when ); in, For the current moment, For the step amplitude, satisfying occupy The percentage does not exceed the preset upper limit to ensure that upper-layer services are not affected.

[0034] Within the preset frequency sweep time interval Within the load, by superimposing a small sinusoidal disturbance through the load modulation unit, the load current can be expressed as: ; in, This represents the amplitude of the frequency sweep disturbance. It changes monotonically within a preset frequency range over time, for example from Scan to The frequency range covers the frequency band near the control loop band.

[0035] By recording the load current time series corresponding to step disturbances and frequency sweep disturbances, a micro-disturbance excitation dataset is formed.

[0036] Step S302: Under the action of the micro-perturbation excitation dataset, the output voltage response waveform, output current response waveform, ripple noise change sequence, electromagnetic radiation spectrum change data, and temperature change time series of key components of the high-frequency switching power supply under test are collected to form an online raw dynamic dataset.

[0037] The logic for obtaining the micro-perturbation excitation dataset is as follows: The amplitude of load step disturbance near the rated load is limited to a preset percentage of the rated load.

[0038] The frequency range of the sweeping sinusoidal disturbance is set to cover the bandwidth of the high-frequency switching power supply control loop and its nearby frequency bands, and the preset duration is maintained at each frequency point.

[0039] The amplitude and frequency distribution of the micro-perturbation excitation dataset are adaptively adjusted based on the observability indices of the output voltage and output current responses.

[0040] Step S303: Time alignment and feature extraction are performed on the online original dynamic dataset. The overshoot, settling time, and oscillation decay characteristics of the output voltage and output current are calculated. The ripple spectrum energy distribution index, electromagnetic radiation spectrum amplitude index, and temperature rise rate index are calculated and combined to form an online dynamic response dataset.

[0041] Collect output voltage time series from online raw dynamic datasets and output current time series And extract the following features from it: Overshoot and steady-state value: The average steady-state output voltage before the step jump is taken as the steady-state value. Calculate the maximum value of the output voltage within a certain period after the step disturbance. Then the voltage overshoot can be defined as .

[0042] Similarly, the output current overshoot can be defined.

[0043] Recovery time: Set the allowable error bandwidth to After the step disturbance, find the first time the output voltage enters... And the time point that does not exceed this range afterward is denoted as... The recovery time is .

[0044] Oscillation decay characteristics: During the step response process, the peak output voltage sequence can be extracted, and the oscillation decay rate can be defined as the ratio of the difference between adjacent peak values. The smaller the oscillation decay rate, the weaker the damping, and the more obvious the corresponding control loop degradation may be.

[0045] Ripple spectral energy index: Perform a discrete Fourier transform on the ripple noise time series to obtain a spectral amplitude sequence. Within a preset frequency band (e.g., near the switching frequency and its harmonics), accumulate the spectral amplitude or power to obtain the spectral energy index of each frequency band, thus forming the ripple spectral feature vector.

[0046] Electromagnetic radiation spectral line amplitude index: The amplitude values ​​corresponding to each calibrated frequency point in the electromagnetic radiation spectrum data are read to form a radiation spectrum amplitude vector, which is used to characterize the effects of switching current changes and magnetic device aging on radiation.

[0047] Temperature rise rate index: The rate of temperature change is calculated from the temperature time series of key components, and the temperature slope is calculated before and after the micro-perturbation excitation.

[0048] By using the above rules, all features are combined in a fixed order to form the feature vector in the online dynamic response dataset.

[0049] Step S4 includes the following sub-steps: Step S401: Input the online steady-state operation dataset and the online dynamic response dataset into the digital twin baseline model, and use the parameter inversion algorithm to iteratively adjust the equivalent series resistance of the electrolytic capacitor, the equivalent inductance value, the leakage inductance of the transformer, the on-resistance of the power device, and the control loop gain in the parameter set of the baseline model.

[0050] Parameter inversion algorithms include: A weighted objective function is constructed with components including the fitting error of the output voltage time series, the fitting error of the output current time series, the fitting error of the ripple spectrum, the fitting error of the electromagnetic radiation spectrum, and the fitting error of the temperature change of key components.

[0051] In each iteration, the equivalent series resistance of the electrolytic capacitor, the equivalent inductance, the leakage inductance of the transformer, the on-resistance of the power device, and the gain of the control loop are adjusted simultaneously based on the gradient of the objective function or the search direction.

[0052] When the weighted objective function is less than the preset convergence threshold or the number of iterations reaches the upper limit, the set of inverted aging sensitive parameters is output.

[0053] In a digital twin baseline model, the baseline model parameter set is denoted as a parameter vector. This includes the equivalent series resistance of electrolytic capacitors, equivalent inductance, transformer leakage inductance, power device on-resistance, and control loop gain. Given a micro-perturbation excitation dataset and an online steady-state operation dataset, the digital twin baseline model can be used to measure parameters. The simulation output voltage time series, output current time series, ripple spectrum characteristics, radiation spectrum characteristics, and temperature change characteristics of key components are shown below.

[0054] Measurement feature vectors are extracted from the online steady-state operation dataset and the online dynamic response dataset, including output voltage time series feature vector, output current time series feature vector, ripple spectrum feature vector, electromagnetic radiation spectrum feature vector, and temperature rise feature vector.

[0055] Digital twin baseline model in parameters The corresponding simulation feature vector is then output.

[0056] Various errors can be defined as the sum of squares or mean squares of the differences in eigenvectors, for example: The output voltage error index is the sum of squares or normalized mean square error of the differences in output voltage characteristics; The output current error index is the sum of squares or normalized mean square error of the differences in output current characteristics. Ripple spectrum error, electromagnetic radiation spectrum error, and temperature rise error are also defined as the sum of squares or root mean square of the differences between the measured eigenvector and the simulated eigenvector, respectively.

[0057] By assigning weights to the aforementioned errors according to their importance, a weighted objective function is constructed. The objective function is a weighted sum of various errors; the smaller the value, the closer the simulation results are to the measured data.

[0058] In step S402, in each iteration, the fitting errors of the output voltage time series, output current time series, ripple spectrum, electromagnetic radiation spectrum, and temperature change of key components obtained from the digital twin baseline model simulation with the online steady-state operation dataset and the online dynamic response dataset are calculated respectively. When the fitting error is lower than the preset threshold, the iteration is stopped, and the inverted aging sensitive parameter set is obtained.

[0059] The following iterative inversion process is adopted: Using the baseline model parameter set as initial parameters Calculate the initial objective function value.

[0060] In each iteration step, the parameter vector is adjusted according to the changing trend of the objective function. The parameter adjustment can be done using gradient descent, quasi-Newton method, or a heuristic method based on the search step size.

[0061] Apply parameter constraints during the iteration process, for example: The equivalent series resistance of the electrolytic capacitor remains positive and is not lower than the reference value when it is a new component. The equivalent inductance and transformer leakage inductance remain positive and do not exceed the upper limit determined by the aging sample dataset; The on-resistance of the power device and the gain of the control loop vary within a preset upper and lower bound. When the objective function falls below the preset convergence threshold or the objective function decreases below the preset change threshold after multiple consecutive iterations, the iteration stops, and the parameter vector at this point is used as the inversion aging sensitive parameter set.

[0062] By inverting the parameters under the above constraints, it can be ensured that the inverted set of aging-sensitive parameters is reasonable in the physical sense of the circuit and is consistent with the online steady-state operation dataset and the online dynamic response dataset in multiple dimensions.

[0063] Step S403: Based on the offset ratio between the inverted aging sensitive parameter set and the baseline model parameter set, calculate the capacitor aging index, transformer aging index, power device fatigue index, and control loop degradation index, combine the indices to form an aging characteristic index set, and output it as the second analysis result.

[0064] Each aging index is calculated based on the degree of deviation between the inverted aging-sensitive parameter set and the baseline model parameter set.

[0065] The reference values ​​for the equivalent series resistance of electrolytic capacitors in the baseline model parameter set, the statistical values ​​of the equivalent series resistance corresponding to the end-of-life stage in the aging sample data set, and the equivalent series resistance in the inverted aging sensitive parameter set are set as follows.

[0066] The capacitor aging index can be represented by the normalized result of the relative drift between the reference value and the termination value: If the statistical value of the equivalent series resistance at the end of the lifespan is greater than the reference value of the equivalent series resistance of the electrolytic capacitor, then the aging range is defined as the period from the reference value of the equivalent series resistance of the electrolytic capacitor to the statistical value of the equivalent series resistance at the end of the lifespan. The equivalent series resistance in the set of aging-sensitive parameters is linearly interpolated to obtain a dimensionless value between 0 and 1, which is used as the capacitor aging index.

[0067] If the equivalent series resistance in the set of inverted aging sensitive parameters is less than the reference value of the equivalent series resistance of the electrolytic capacitor, the capacitor aging index can be limited to 0. When the equivalent series resistance in the set of inverted aging sensitive parameters exceeds the statistical value of the equivalent series resistance at the end of life stage, the capacitor aging index can be limited to 1.

[0068] The baseline model parameter set includes the transformer leakage inductance reference value, the leakage inductance statistical value corresponding to the end of life stage, and the leakage inductance obtained by inversion.

[0069] In a similar manner, the leakage inductance obtained by inversion is normalized within a linear or piecewise linear interval to obtain the transformer aging index.

[0070] Set reference values ​​for the on-resistance of power devices, statistical values ​​for the end-of-life stage, and inversion values.

[0071] Based on the degree of relative increase in on-resistance, the inverted value is normalized with the reference value of on-resistance of the power device and the statistical value of the end-of-life stage to obtain the fatigue index of the power device.

[0072] Set the control loop gain reference value, end-of-life stage statistics value, and inversion value.

[0073] Based on the degree of attenuation of the control loop gain, the gain change is normalized to the control loop degradation index.

[0074] The above four indices are all normalized to the range of 0 to 1 and combined to form a set of aging characteristic indicators, which are used for subsequent lifespan prediction models and aging degree level determination.

[0075] Step S5 includes the following sub-steps: Step S501: Input the aging characteristic index set in the second analysis result into the lifetime prediction model trained based on the aging sample dataset. Calculate the estimated remaining lifetime of the high-frequency switching power supply under test based on the mapping relationship between capacitor aging index, transformer aging index, power device fatigue index, control loop degradation index and aging stage.

[0076] A regression model trained on an aging sample dataset is used to map the set of aging feature indicators to estimates of remaining lifespan.

[0077] The input variables are four dimensionless indices from the aging characteristic index set: capacitor aging index, transformer aging index, power device fatigue index, and control loop degradation index. The output variable is the estimated remaining lifespan, which can be expressed in hours or days.

[0078] Multiple known aging stage samples are obtained from the aging sample dataset. Each sample has a set of aging characteristic indicators and corresponding actual operating life and remaining life.

[0079] The above samples are used to train a multivariate regression model or a nonlinear fitting model, mapping the four aging indices to lifespan depletion, and then using the lifespan depletion to estimate the remaining lifespan: A lifespan consumption rate of 0 indicates a new component status. A lifetime consumption level of 1 indicates the end of the lifetime. The remaining lifespan is the nominal lifespan multiplied by the remaining lifespan percentage.

[0080] The lifetime prediction model obtained through training can take the aging characteristic index set in the second analysis result as input during the testing phase, and output the lifetime consumption and remaining lifetime estimate of the high-frequency switching power supply under test, and finally form the aging assessment result set.

[0081] Step S502: Based on the aging characteristic index set and the remaining life estimate, the high-frequency switching power supply under test is divided into multiple aging levels according to the preset aging level threshold, and an aging assessment result set is generated. The aging assessment result set includes the aging level, aging percentage and remaining life estimate.

[0082] The logic for determining the level of aging is as follows: The aging level is preset, and each level corresponds to the threshold range of capacitor aging index, transformer aging index, power device fatigue index and control loop degradation index.

[0083] Based on the position of each index in the aging characteristic index set within the corresponding threshold range, a weighted comprehensive score is calculated, and the comprehensive score is mapped to the corresponding aging level.

[0084] When any single index exceeds its highest level threshold, the high-frequency switching power supply under test will be directly determined to be at the highest aging level and marked as a high-risk object in the aging assessment results set.

[0085] Step S503: In rack application scenarios, the aging assessment result sets of multiple high-frequency switching power supplies under test are compared horizontally to generate rack-level aging risk distribution data. Based on the rack-level aging risk distribution data, the load scheduling strategy and replacement recommendations are determined and input to the control terminal.

[0086] The logic for generating the load scheduling strategy is as follows: Based on rack-level aging risk distribution data, health priorities are assigned to each high-frequency switching power supply. Power supplies with low aging levels and high remaining life estimates are classified as high priority, while power supplies with high aging levels or low remaining life estimates are classified as low priority.

[0087] While meeting the total rack power requirements, more load is allocated to high-priority power supplies, while the load on low-priority power supplies is limited to below a preset upper limit.

[0088] When the estimated remaining life of a certain high-frequency switching power supply in the aging assessment results is lower than the preset maintenance time window, it is marked as a replacement object in the load scheduling strategy, and a corresponding planned replacement suggestion is generated.

[0089] Compared with existing schemes that rely on steady-state voltage and current and offline sampling detection, the present invention has the following advantages: This method applies a controlled-amplitude micro-perturbation excitation near the rated load and collects data within a preset test window. The perturbation amplitude is limited to a certain proportion of the rated load, which neither causes interruption of upper-layer services nor affects the dynamic characteristics of the control loop and the implicit aging characteristics of the power stage. This allows the aging effect to be amplified and observed in the dynamic response, thus enabling online aging tests during the operation of the data center.

[0090] This method simultaneously acquires output voltage time series, output current time series, ripple spectrum characteristics, electromagnetic radiation spectrum characteristics, and temperature changes of key components. By extracting dynamic response characteristics such as overshoot, synchrotron time, oscillation decay characteristics, band energy indicators, and temperature rise rate, multi-source information is input into a digital twin baseline model. Through parameter inversion, aging-sensitive parameters such as the equivalent series resistance of electrolytic capacitors, transformer leakage inductance, power device on-resistance, and control loop gain are obtained, and their deviation is normalized into capacitor aging index, transformer aging index, power device fatigue index, and control loop degradation index. This enables aging identification at the device level and mechanism level, rather than just a coarse-grained judgment of whether the entire system is good or bad.

[0091] This method constructs an aging sample dataset through accelerated aging tests during the production or experimental phase, calibrates aging-sensitive parameters and actual lifespan information for different aging stages, and establishes a mapping relationship between aging stages and a lifespan prediction model. In field testing, the inverted aging characteristic index set is input into the lifespan prediction model to obtain lifespan depletion and remaining lifespan estimates. This establishes a quantifiable and consistent correspondence between online field test results and factory test data, improving the reliability and interpretability of remaining lifespan predictions.

[0092] This method not only targets the aging level and remaining lifespan of a single high-frequency switching power supply, but also performs polling tests on multiple power modules within the same rack. It then compares the aging assessment results of each module horizontally to generate rack-level aging risk distribution data. Combined with preset power demand constraints and maintenance strategies, this method can automatically generate load scheduling strategies, allocating more load to high-priority power supplies with lower aging levels and longer remaining lifespans, while keeping low-priority power supplies with higher aging levels at lower load levels. It also provides planned replacement recommendations for modules nearing the end of their lifespan, reducing the risk of sudden failures and improving the overall availability and economy of the power supply system.

[0093] Because this method is based on a digital twin model with a clear physical mechanism, it restores the multi-source dynamic response data to the changes in aging-sensitive parameters inside the circuit through parameter inversion, and then maps the parameter offset to the aging index and lifespan consumption. The entire link maintains a consistent physical meaning and mathematical rules from measurement data to evaluation results, making the aging evaluation results easy for engineers to understand and track. It is also beneficial to combine actual operation and maintenance experience to optimize the model and threshold settings, forming a closed-loop iteration.

[0094] In summary, this invention integrates micro-disturbance excitation, multi-source data acquisition, digital twin parameter inversion, and lifetime prediction models under non-stop conditions, realizing device-level aging degree testing and refined remaining lifetime assessment of high-frequency switching power supplies. The results are then extended to rack-level load scheduling and operation and maintenance decisions, significantly improving the reliability and refined management level of power supply systems in communication equipment rooms and data centers.

[0095] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read-Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. 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.

[0096] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for testing the aging degree of a high-frequency switching power supply, characterized in that, Comprising the following steps: Step S1, retrieve high-frequency switching power supply aging sample data from the aging sample database, establish a digital twin baseline model containing aging sensitive parameters and output the first analysis result; Step S2, collect online steady-state operation data of the to-be-tested high-frequency switching power supply in the target application scenario and pre-process to form an online steady-state operation data set; Step S3, apply a small perturbation excitation to the to-be-tested high-frequency switching power supply within a preset test time window, collect dynamic response operation data and form an online dynamic response data set; Step S4, input the online steady-state operation data set and the online dynamic response data set into the digital twin baseline model, obtain the aging sensitive parameters through parameter inversion and generate an aging characteristic index set, and output the second analysis result; Step S5, evaluate the aging degree and remaining life of the to-be-tested high-frequency switching power supply according to the first analysis result and the second analysis result, generate an aging evaluation result set and a load scheduling strategy and input to the control end.

2. The method of claim 1, wherein the high-frequency switching power supply is a switching power supply for a mobile communication terminal. The step S1 comprises the following sub-steps: Step S101, retrieve high-frequency switching power supply aging sample data from the aging sample database, the aging sample data comprising output voltage time series, output current time series, ripple noise time series, electromagnetic radiation spectrum data and key device temperature time series under multiple aging stages, and corresponding aging stage labels, forming an aging sample data set; Step S102, establish a parameterized digital twin baseline model based on the high-frequency switching power supply circuit topology, the parameters of the digital twin baseline model including equivalent series resistance of electrolytic capacitor, equivalent inductance value, transformer leakage inductance, power device on-resistance and control loop gain, forming a baseline model parameter set; Step S103, input the aging sample data set into the digital twin baseline model, calculate the aging sensitive parameter values corresponding to each aging stage by fitting the consistency of the simulation waveforms and simulation spectrum under each aging stage with the aging sample data set, establish a mapping relationship between the aging sensitive parameters and the aging stages, and output the first analysis result.

3. The method of claim 1, wherein the high-frequency switching power supply is a switching mode power supply (SMPS). The step S2 comprises the following sub-steps: Step S201, connect the to-be-tested high-frequency switching power supply to the load system of the target application scenario, and collect output voltage time series, output current time series, ripple noise time series, electromagnetic radiation spectrum data and key device temperature time series online through voltage and current sampling circuit, ripple acquisition circuit, electromagnetic radiation probe and temperature sensor, forming an online original steady-state data set; Step S202, perform time reference alignment, outlier rejection and noise filtering on the online original steady-state data set, combine the aligned output voltage time series, output current time series, ripple noise time series, electromagnetic radiation spectrum data and key device temperature time series to form an online steady-state operation data set.

4. The method of claim 1, wherein the high-frequency switching power supply is a switching power supply for a mobile communication terminal. The step S3 comprises the following sub-steps: Step S301, within a preset test time window, apply small amplitude step perturbation and sweep frequency sinusoidal perturbation to the load current near the rated load of the to-be-tested high-frequency switching power supply through the load modulation unit, record the perturbation signal time series, and form a small perturbation excitation data set; Step S302, under the action of the micro-disturbance excitation data set, the output voltage response waveform, the output current response waveform, the ripple noise change sequence, the electromagnetic radiation spectrum change data and the key device temperature change time sequence of the to-be-tested high-frequency switching power supply are collected to form an online original dynamic data set; Step S303, time alignment and feature extraction are performed on the online original dynamic data set, the overshoot, the return-to-steady time and the oscillation decay characteristics of the output voltage and the output current are calculated, the ripple spectrum energy distribution index, the electromagnetic radiation spectrum line amplitude index and the temperature rise rate index are calculated, and an online dynamic response data set is formed.

5. The method of claim 1, wherein the high-frequency switching power supply is a switching power supply for a mobile communication terminal. The step S4 includes the following sub-steps: Step S401, inputting the online steady-state operation data set and the online dynamic response data set into the digital twin baseline model, and iteratively adjusting the equivalent series resistance of the electrolytic capacitor, the equivalent inductance value, the transformer leakage inductance, the power device on-resistance and the control loop gain in the baseline model parameter set by using a parameter inversion algorithm; Step S402, in each iteration, the fitting error of the output voltage time sequence, the output current time sequence, the ripple spectrum, the electromagnetic radiation spectrum and the key device temperature change simulated by the digital twin baseline model are calculated respectively, and the fitting error is compared with the online steady-state operation data set and the online dynamic response data set; when the fitting error is lower than a preset threshold, the iteration is stopped, and an inversion aging sensitive parameter set is obtained; Step S403, according to the offset proportion of the inversion aging sensitive parameter set and the baseline model parameter set, a capacitor aging index, a transformer aging index, a power device fatigue index and a control loop degradation index are calculated, the indexes are combined to form an aging characteristic index set, and the aging characteristic index set is output as a second analysis result.

6. The method of claim 1, wherein the high-frequency switching power supply is a switching power supply for a mobile communication terminal. The step S5 includes the following sub-steps: Step S501, inputting the aging characteristic index set in the second analysis result into a life prediction model trained based on an aging sample data set, and calculating a residual life estimate value of the to-be-tested high-frequency switching power supply according to the mapping relationship between the capacitor aging index, the transformer aging index, the power device fatigue index and the control loop degradation index and the aging stage; Step S502, based on the aging characteristic index set and the residual life estimate value, the to-be-tested high-frequency switching power supply is divided into multiple aging levels according to a preset aging degree level threshold, and an aging evaluation result set is generated, which includes the aging degree level, the aging percentage and the residual life estimate value; Step S503, in a rack application scenario, the aging evaluation result sets of multiple to-be-tested high-frequency switching power supplies are compared horizontally to generate a rack-level aging risk distribution data, a load scheduling strategy and a replacement suggestion are determined according to the rack-level aging risk distribution data, and the load scheduling strategy and the replacement suggestion are input to a control end.

7. The method of claim 4, wherein the step of applying a stress to the high frequency switching power supply comprises applying a stress to the high frequency switching power supply by applying a voltage to the high frequency switching power supply. The acquisition logic of the micro-disturbance excitation data set is as follows: The amplitude of the load step disturbance is limited to be less than a preset percentage of the rated load near the rated load; The frequency range of the sweep sine disturbance is set to cover the control loop bandwidth of the high-frequency switching power supply and the frequency band near it, and each frequency point is maintained for a preset duration; The amplitude and frequency distribution of the micro-disturbance excitation data set are adaptively adjusted according to the observability index of the output voltage and output current response.

8. The method of claim 5, wherein the high-frequency switching power supply is a switching power supply for a mobile communication terminal. The parameter inversion algorithm comprises: a weighted objective function is constructed with output voltage time series fitting error, output current time series fitting error, ripple spectrum fitting error, electromagnetic radiation spectrum fitting error and key device temperature change fitting error as components; in each iteration, the equivalent series resistance of electrolytic capacitor, equivalent inductance value, transformer leakage inductance, power device on-resistance and control loop gain are adjusted simultaneously according to the gradient of the objective function or the search direction; when the weighted objective function is less than the preset convergence threshold or the number of iterations reaches the upper limit, the set of inversion aging sensitive parameters is output.

9. The method for testing the aging degree of a high-frequency switching power supply as described in claim 6, characterized in that, The determination logic of the aging degree level is: preset aging degree levels, each level corresponding to threshold intervals of capacitor aging index, transformer aging index, power device fatigue index and control loop degradation index; according to the position of each index in the corresponding threshold interval in the aging characteristic index set, the weighted comprehensive score is calculated, and the comprehensive score is mapped to the corresponding aging degree level; when any single index exceeds the highest level threshold, the high-frequency switching power supply to be tested is directly determined as the highest aging degree level and marked as a high-risk object in the aging evaluation result set.

10. The method of claim 6, wherein the high frequency switching power supply is a switching power supply for a mobile communication terminal. The generation logic of the load scheduling strategy is: according to the rack-level aging risk distribution data, health priorities are assigned to each high-frequency switching power supply, the power supply with low aging degree level and high remaining life estimation value is divided into high priority, and the power supply with high aging degree level or low remaining life estimation value is divided into low priority; under the condition of meeting the rack total power demand constraint, more loads are allocated to high-priority power supplies, and the load of low-priority power supplies is limited below the preset upper limit; when the remaining life estimation value of a high-frequency switching power supply in the aging evaluation result set is lower than the preset maintenance time window, it is marked as a replacement object in the load scheduling strategy, and a corresponding planned replacement suggestion is generated.

Images