A gallium nitride charger circuit fault diagnosis method and system

Abstract

The application relates to the technical field of charger health management, and discloses a gallium nitride charger circuit fault diagnosis method and system, the method comprising the following steps: acquiring a conduction resistance time sequence and device junction temperature data, and synchronously mapping to obtain heat accumulation and a temperature slope; performing period-by-period difference on the temperature slope to obtain an attenuation factor, and performing weighted smoothing processing to obtain a degradation rate estimation value; arranging the degradation rate estimation value into a degradation trajectory sequence and performing logarithmic regression to obtain a degradation trend prediction value; if a preset safety threshold is exceeded, determining a degradation rate deviation and calculating a remaining safety working time, and if a safety time lower limit is lower than the safety time lower limit, triggering a risk early warning. The method can quantize the remaining safety working time of a gallium nitride transistor.

Description

Technical Field

[0001] This invention relates to the field of charger health management technology, and in particular to a method and system for diagnosing faults in gallium nitride charger circuits. Background Technology

[0002] In high-power charger applications, gallium nitride (GaN) transistors undergo physical degradation during prolonged full-load operation, resulting in a continuous increase in on-resistance. The degradation mechanism involves hot carrier injection leading to gate interface state trapping. The accumulation of these trapped states causes depletion of the two-dimensional electron gas (2DEG) channel, further contributing to the increased on-resistance. This increased on-resistance generates more heat under the same current, raising the junction temperature, which in turn accelerates material aging, creating a positive feedback loop between increasing on-resistance and heat accumulation. Pre-diagnosis and health management technologies require monitoring the degradation rate to predict the remaining usable lifespan of the device. However, this degradation process is not constant; it exhibits a non-linear accelerating characteristic with heat accumulation.

[0003] Existing technologies rely on preset fixed temperature or resistance thresholds for fault diagnosis and real-time monitoring of device operating parameters. When these parameters exceed the preset thresholds, protective actions are triggered. However, fixed thresholds represent static boundaries, responding only when device parameters have clearly exceeded these limits. By this time, the device is often nearing its failure threshold, failing to provide early warnings during the accelerated degradation phase. More importantly, fixed threshold mechanisms cannot recognize the dynamic evolution of degradation rates accelerated by heat accumulation, cannot determine whether the current degradation rate is within the normal aging range or has entered the accelerated failure phase, and cannot quantify the device's safety margin from the failure threshold or provide remaining safe operating time based on the current degradation trajectory. There is a degradation assessment scheme for battery capacity retention rate in the prior art (CN202011134267.1, IPC classification number: G01R31 / 367), which judges the risk of capacity drop by calculating the slope change of capacity retention rate with the number of cycles. However, this scheme is geared towards the detection of abrupt changes in macroscopic performance indicators and is not applicable to the degradation process of gallium nitride transistor on-resistance which is continuous and gradual and has an accelerating trend. It cannot track the evolution trend of the degradation rate itself and its dynamic coupling relationship with heat accumulation.

[0004] In summary, existing technologies cannot quantify the remaining safe operating time of gallium nitride transistors. Summary of the Invention

[0005] This invention provides a method and system for diagnosing gallium nitride (GaN) charger circuit faults, in order to solve the problem of being unable to quantify the remaining safe operating time of GaN transistors.

[0006] In a first aspect, to solve the above-mentioned technical problems, the present invention provides a gallium nitride charger circuit fault diagnosis method, comprising: The on-resistance time series and junction temperature sensing data of gallium nitride transistors under full load conditions are obtained and synchronously mapped to obtain the heat accumulation and temperature slope. The temperature slope is differentially analyzed periodically to obtain the attenuation factor. The attenuation factor is then used to perform weighted smoothing on the temperature slope and the heat accumulation to obtain an estimated degradation rate. Arrange the degradation rate estimates in chronological order to obtain a degradation trajectory sequence. Perform logarithmic regression on the degradation trajectory sequence to obtain the degradation trend prediction and trajectory change slope. If the predicted degradation trend value exceeds the preset safety threshold, the degradation rate deviation is determined based on the trajectory change slope. Based on the degradation rate deviation and the trajectory change slope, the remaining safe working time is calculated based on the preset on-resistance limit value. If the remaining safe working time is lower than the preset safety time lower limit, a risk warning is triggered.

[0007] In a second aspect, the present invention provides a gallium nitride charger circuit fault diagnosis system, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the gallium nitride charger circuit fault diagnosis method as described above.

[0008] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the gallium nitride charger circuit fault diagnosis method as described above.

[0009] Compared with the prior art, the present invention has the following beneficial effects: (1) The present invention obtains the comprehensive degradation rate by weighted fusion of the degradation rate deviation and the trajectory change slope, and calculates the remaining safe working time based on the preset on-resistance limit value, thereby realizing the quantitative assessment of degradation risk and solving the problem that the existing fixed threshold mechanism cannot quantify the remaining safe working time of the device.

[0010] (2) The present invention obtains the attenuation factor by periodically differentiating the temperature slope, and uses the attenuation factor to perform weighted smoothing of the temperature slope and the amount of heat accumulation to obtain the degradation rate estimate. It can identify the dynamic evolution mode of degradation rate with heat accumulation and distinguish between normal aging and accelerated failure stages.

[0011] (3) This invention obtains the heat accumulation and temperature slope by synchronously mapping the on-resistance time series and the device junction temperature sensing data, and establishes a dynamic correspondence between electrical parameter degradation and thermal response, providing a coupled data basis for degradation rate estimation. Attached Figure Description

[0012] Figure 1 This is a flowchart illustrating a gallium nitride charger circuit fault diagnosis method provided in an embodiment of the present invention. Detailed Implementation

[0013] 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.

[0014] Reference Figure 1 The first embodiment of the present invention provides a method for diagnosing faults in a gallium nitride charger circuit, comprising the following steps: In step S11, the on-resistance time series and junction temperature sensing data of the gallium nitride transistor under full load are acquired and synchronously mapped to obtain the heat accumulation and temperature slope, including: The on-resistance time series is subjected to sliding window mid-range filtering to obtain a smoothed on-resistance series; The smooth on-resistance sequence and the device junction temperature sensing data are timestamped to obtain a synchronization mapping matrix; Based on the synchronization mapping matrix, the smooth on-resistance sequence is integrated over time to obtain the accumulated heat, and the junction temperature sensing data of the device is differentiated over time to obtain the temperature slope.

[0015] It should be noted that under full load conditions, with a current load current of 20A and a switching frequency of 100kHz, a sliding time window algorithm is used to determine the duration of the gallium nitride transistor in a stable conduction state. The sliding time window width is set to 20 sampling points (corresponding to 20μs), and the sliding step size is 5 sampling points (corresponding to 5μs). The device is considered to have entered a stable conduction state when the drive voltage is greater than 4.5V for 15 consecutive sampling points within the window. This judgment condition is based on the premise that the time for the drive voltage to stabilize to the conduction plateau after the switch is completed does not exceed 5μs, and 15 consecutive points correspond to 15μs, covering the transient margin. Based on this duration, the sampling window is set to the middle 50% of the conduction cycle to avoid ringing interference and Miller plateau at the moment of switching. The first 10% and last 10% of the switching moment have voltage overshoot and current oscillation, while the middle 50% corresponds to the steady-state range where the device is fully turned on and the drain-source voltage is the lowest. Within the sampling window, the gate drive voltage of the gallium nitride transistor is acquired via high-speed analog-to-digital conversion. The analog-to-digital converter has a resolution of 12 bits, an input voltage range of 0 to 10V, and a corresponding quantization accuracy of 2.44mV. The sampling rate is consistent with the on-resistance acquisition at 1MHz, forming a drive voltage sequence. The preset turn-on threshold is 4.5V. This threshold is set based on the fact that the typical gate threshold voltage range of gallium nitride enhancement-mode transistors is 1.5V to 3V. 4.5V exceeds the maximum threshold voltage by more than 50%, ensuring that the device is fully in the low-resistance conduction region. This threshold is derived from the transfer characteristic curve in the device datasheet. When a voltage value in the drive voltage sequence is detected to be greater than the preset turn-on threshold of 4.5V, it indicates that the device has fully entered the conduction state, at which point the junction temperature extraction mechanism is triggered.

[0016] Specifically, when extracting the junction temperature, the current drive voltage and drain-source current data are input, and the junction temperature is estimated using a thermal network model lookup table method, outputting the current internal junction temperature of the device. A thermal network model lookup table is pre-established, using a third-order Foster network topology. The thermal resistance parameters are Rth1=0.5K / W, Rth2=0.3K / W, Rth3=0.2K / W, and the thermal capacity parameters are Cth1=0.02J / K, Cth2=0.15J / K, Cth3=1.2J / K. These parameters are obtained through multi-exponential fitting of the device's thermal impedance curve, with a fitting error of less than 5%. The current power P=Vds×Id is calculated based on the drive voltage and drain-source current. Using the power and current ambient temperature as input, the temperatures of the three thermal network nodes T1, T2, and T3 are obtained from the lookup table, and the junction temperature Tj=T1+T2+T3. The ambient temperature is acquired through an on-chip temperature sensor, with an initial ambient temperature of 25℃. The table has a temperature resolution of 1℃ and a power resolution of 0.5W. The table storage space is approximately 4KB, which is fully compatible with the MCU's Flash capacity.

[0017] After the junction temperature is extracted, the resistive compensation coefficient at the current temperature is calculated using a pre-calibrated temperature impedance polynomial fitting model. The calibration procedure for the temperature impedance polynomial model involves placing gallium nitride transistors from the same batch in a constant temperature chamber and measuring the on-resistance at 26 temperature points in 5°C increments within a range of 25°C to 150°C. The resistance is recorded after holding each temperature point at a constant temperature for 5 minutes. The resistance value is then calculated using the third-order polynomial Rds(T) = a0 + a1T + a2T. 2 +a3T 3 Least squares fitting is performed, and the resulting coefficients a0 to a3 are the calibration parameters. For example, after calibration, a0 = 128.5, a1 = -0.42, a2 ​​= 0.0068, and a3 = -1.2 × 10⁻⁶. -5 Goodness of fit R 2 Greater than 0.99. Multiply the original acquired resistance fluctuation data by this compensation coefficient to obtain the compensated resistance fluctuation. This compensation process eliminates the interference of thermal effects on the device resistance.

[0018] For example, at a certain sampling moment, the current junction temperature is estimated to be 85℃ by looking up a table using the thermal network. The temperature impedance polynomial model outputs a compensation coefficient of 0.92 at this temperature, and the original resistance fluctuation measurement is 45mΩ. Multiplying this by the compensation coefficient yields a compensation value of 41.4mΩ. Based on the compensated resistance fluctuation, the drain-source voltage and load current of the gallium nitride transistor are periodically sampled at a high frequency of 1MHz, and the on-resistance is calculated in real time to obtain a continuous original sequence of on-resistance. The 1MHz sampling rate is set based on a 100kHz switching frequency, with 10 data points sampled per cycle, satisfying the Nyquist sampling theorem's requirement for capturing slowly changing on-resistance signals. Simultaneously, the microcontroller's analog-to-digital converter can complete a single sampling within 1μs. Noise suppression is achieved by applying a sliding window median filter to the original on-resistance sequence. The sliding window width is set to 11 sampling points, determined based on the spike characteristics of the ambient noise. The typical pulse width of spike noise does not exceed 5 sampling points (corresponding to 5μs). The 11-point window ensures that at least 6 normal sampling points constitute the majority, and median filtering can completely eliminate spike noise with pulse widths less than half the window width. For the first and last 5 sampling points of the sequence, median filtering is performed after expanding the data using a mirror extension method. The implementation of median filtering on the MCU only requires maintaining a circular buffer of length 11 and performing insertion sort to obtain the median. The computational complexity of a single filtering operation is O(W), i.e., 11 comparison operations. For example, if there is a random spike noise with an amplitude of 50mΩ in the original on-resistance sequence at a certain sampling moment, the output after 11-point sliding window median filtering is 152mΩ, which approximates the actual physical resistance of the device under this condition of 150mΩ.

[0019] It should be noted that the sampling rate of the junction temperature sensing data of gallium nitride transistors is lower than that of the electrical parameters. The junction temperature sensing data is acquired by an on-chip temperature sensor or an external thermistor, with a nominal analog-to-digital conversion rate of 10kHz, corresponding to a sampling period of 100μs. The smoothed on-resistance sequence sampling rate is 1MHz, corresponding to a sampling period of 1μs. The time bases of the two sets of data differ by two orders of magnitude, requiring timestamp alignment. A linear interpolation algorithm is used to resample the 10kHz junction temperature data, inserting 99 intermediate values ​​with a 1μs step between adjacent junction temperature sampling points, ensuring strict alignment between the junction temperature data and the 1MHz resistance sequence on a μs-scale. The linear interpolation is selected based on the condition that the change in junction temperature within a 100μs interval does not exceed 0.3℃; the error introduced by linear interpolation is negligible, and higher-order interpolation methods are not required. The aligned timestamps, resistance values, and temperature values ​​are assembled into a three-column matrix, with each row corresponding to the same moment, containing time, resistance, and temperature dimensions, resulting in a synchronization mapping matrix.

[0020] For example, at time 10:05:20.000000, the resistance value is 152 mΩ, and the original junction temperature is 84.8℃; at the next junction temperature sampling time 10:05:20.000100, the resistance value is 153 mΩ, and the original junction temperature is 84.9℃. After linear interpolation, the corresponding junction temperature interpolation values ​​of 84.801℃, 84.802℃, up to 84.899℃ are obtained for the intermediate 99 time points, which are strictly aligned with the 1MHz resistance data. All data points are combined to form a synchronization mapping matrix.

[0021] Specifically, based on the constructed synchronization mapping matrix, a smoothed on-resistance sequence within a set time window is extracted, and a trapezoidal integral algorithm is used for time integration. The time window is set to 10ms, selected based on the fact that a 100kHz switching frequency corresponds to 1000 switching cycles. The 10ms window contains 10,000 resistance sampling data points, and the statistical sample size meets the resolution requirements for heat accumulation calculation. Simultaneously, the time window should not be too large to avoid smoothing out short-term fluctuations in the degradation trend. The trapezoidal integral calculation formula is a weighted sum of the resistance averages of adjacent sampling points. That is, the arithmetic mean of the resistance values ​​at each adjacent sampling time is multiplied by the square of the load current and then by the sampling interval. The load current is 20A, and the sampling interval Δt is 1μs. The heat accumulation is obtained by accumulating the integral elements of all 10,000 adjacent point pairs within the 10ms window. The accuracy of the trapezoidal integral is based on the condition that the 1MHz sampling rate is much higher than the bandwidth of the slowly varying on-resistance signal. If the bandwidth of this signal is less than 1kHz, the trapezoidal integral error is O(Δt). 2 It is superior to O(Δt) of rectangular integral.

[0022] It should be noted that, simultaneously, the junction temperature sensing data from the synchronization mapping matrix is ​​extracted, and the center difference algorithm is used for time-domain differential calculation. The center difference calculation formula is the temperature difference between adjacent sampling points divided by twice the sampling interval, i.e., dT / dt=[T(i+1)-T(i-1)] / (2Δt), where Δt is 100μs, corresponding to the junction temperature physical sampling period. The center difference is selected because, compared with forward and backward differences, it provides more accurate estimation of slowly varying signals. It should be noted that the effective time resolution of the data after linear interpolation resampling still depends on the junction temperature physical sampling period of 100μs, and the actual derivative information of the differential operation is limited by the physical sampling interval. A two-dimensional feature space is constructed with the calculated heat accumulation as the abscissa and the temperature slope as the ordinate. The real-time calculated numerical pairs are mapped to this two-dimensional feature space to form discrete data points. For multiple data points within a continuous time period, considering the computing power constraints of the MCU, a threshold-based two-dimensional feature space partitioning algorithm is used for degenerate feature extraction. First, the heat accumulation and temperature slope dimensions are normalized to the 0-1 range using both maximum and minimum methods to eliminate dimensional differences. Then, a pre-defined grid is used in the normalized two-dimensional space, with a grid step size of 0.1, dividing the space into 10×10 grid cells (100 cells in total). The data point density within each grid cell is calculated, and the continuous grid region with the highest data point density is determined as the main trajectory region. A 4-connectivity rule is applied, meaning only adjacent grid cells (up, down, left, and right) are considered connected. After smoothing the data points within the main trajectory region using a moving average, the real-time correspondence between heat accumulation and temperature slope is output.

[0023] For example, within a 10ms time window, the trapezoidal integral accumulates the integral elements of 10,000 sampling points, yielding a cumulative heat of 0.608J. Three adjacent junction temperature sampling points are 84.6℃, 84.8℃, and 85.0℃, with an adjacent sampling interval of 100μs. The central difference is dT / dt=(85.0-84.6) / (2×100×10⁻¹⁰). -6=2000℃ / s, meaning the temperature slope at that moment is 2000℃ / s. In the two-dimensional feature space, as the accumulated heat increases from 0.3J to 0.6J, the main trajectory region extracted by the partitioning algorithm shows a trend temperature slope that non-linearly increases from 50℃ / s to 100℃ / s, forming a real-time correspondence between the accumulated heat and the temperature slope under full-load conditions. The values ​​from 50 to 100℃ / s are trend values ​​smoothed by the partitioning algorithm, belonging to different time scales than the instantaneous temperature slope of 2000℃ / s obtained by the central difference calculation in S11. It should be noted that the temperature slope and heat accumulation output from S11 are normalized before entering S12. The temperature slope is divided by the rated temperature slope reference value under the current operating condition to obtain the dimensionless temperature slope. The rated temperature slope reference value is 2000℃ / s, which is the typical temperature change rate of a healthy device in the stable conduction range under full load conditions. This value is determined by the statistical average of measured values ​​from newly manufactured devices in the same batch under standard full load conditions. The normalized temperature slope value is between 0.02 and 3. The heat accumulation is transferred to S12 window by window using the integral value of each 10ms window. The gradient calculation uses the increment between consecutive windows. The Q value increases monotonically with running time and is only cleared to zero when the device is replaced or the system is reset. It should be noted that the normalized temperature slope entering S12 EWMA is the trend value after smoothing by the partitioning algorithm, not the instantaneous value of the center difference. For example, the trend temperature slope of the main trajectory region extracted by the partitioning algorithm is 100℃ / s. After dividing by the normalized baseline value of 2000℃ / s, the normalized temperature slope is obtained as 100 / 2000=0.05. This normalized trend value is used as the input of S12 EWMA.

[0024] In step S12, the temperature slope is differentially analyzed periodically to obtain an attenuation factor. This attenuation factor is then used to perform a weighted smoothing process on the temperature slope and the accumulated heat to obtain an estimated degradation rate, including: Calculate the slope difference of the temperature slope in adjacent switching cycles to obtain the cycle-by-cycle change; The attenuation factor is determined by comparing the periodic change with a preset change threshold. The temperature slope is obtained by performing an exponentially weighted moving average using the attenuation factor; The rate of change of the temperature slope sequence relative to the accumulated heat is calculated to obtain an estimate of the degradation rate.

[0025] Specifically, a first-order forward differencing algorithm is used to process the temperature slope sequence obtained in the previous stage. The input consists of a temperature slope sequence containing multiple consecutive switching cycles. The slope difference between adjacent cycles is calculated, and the cycle-by-cycle change is output. The sampling update period for the temperature slope sequence is 100 μs, consistent with the junction temperature physical sampling period. The cycle-by-cycle change is calculated using first-order forward differencing during adjacent sampling update cycles. It should be noted that adjacent switching cycles refer to the slope difference between the endpoints of two representative switching cycles within the 100 μs temperature slope sampling update period. The temperature slope within one update interval is the difference calculated at the beginning of that interval and is retained until the next update time. This cycle-by-cycle change characterizes the fluctuation amplitude of the thermal response rate of the gallium nitride transistor under continuous operation. A larger change indicates more severe fluctuations in the thermal response rate, and a greater interference of thermal disturbances on the degradation rate estimation, requiring a larger attenuation weight to suppress transient thermal disturbances. Conversely, a smaller change indicates that the current process is in a steady state, and more current information should be retained to accurately track the degradation trend.

[0026] For example, if the temperature slopes of two adjacent cycles in the input sequence are 2.5 and 3.1, respectively, the output cycle-by-cycle change is 0.6 after a first-order forward difference operation.

[0027] It should be noted that the cycle-by-cycle change reflects the instantaneous fluctuation of the device's thermal response. The preset change threshold is set to 0.5. This threshold is based on the cycle-by-cycle change data of 20 gallium nitride transistors from the same batch, continuously monitored for 100 hours, and the distribution characteristics of the normalized temperature slope's cycle-by-cycle change were statistically analyzed, with the 75th quantile used as the threshold. Here, "cycle" corresponds to a 100μs sampling update cycle for the temperature slope. When the cycle-by-cycle change exceeds the preset threshold of 0.5, the current cycle is considered to be subject to a local transient thermal disturbance, and an attenuation factor of 0.2 is assigned. When the cycle-by-cycle change is less than or equal to the preset threshold of 0.5, the current cycle is considered to be in a steady-state thermal accumulation process, and an attenuation factor of 0.8 is assigned. To avoid frequent switching of the attenuation factor near the threshold causing jitter in the smooth output, a hysteresis mechanism is introduced. The attenuation factor switching is only executed after the cycle-by-cycle change crosses the threshold and maintains the same crossing direction for three consecutive cycles; a single-cycle instantaneous crossover does not trigger switching. The values ​​of the first attenuation factor 0.2 and the second attenuation factor 0.8 are determined by performing a grid search on the combination of attenuation factors in the range of 0.1 to 0.9 with a step size of 0.1. On the accelerated life test data, the correlation coefficient between the estimated degradation rate and the actual degradation trend is used as the optimization objective. The combination of 0.2 and 0.8 achieves the optimal balance between suppressing noise and preserving the trend.

[0028] For example, the calculated periodic change of 0.6 is greater than the preset change threshold of 0.5, so it is determined that the period corresponds to a local transient thermal disturbance, and the output attenuation factor is 0.2.

[0029] The temperature slope and the accumulated heat are subjected to an exponentially weighted moving average using the decay factor to obtain a temperature slope sequence and a smoothed accumulated heat sequence. The weighted smoothing of the accumulated heat is performed using the same exponentially weighted moving average iterative formula and decay factor sequence as the temperature slope, with the iterative formula being Q. n '=α n ·Q n +(1-α n )·Q n-1 ', where Q n Q represents the cumulative heat amount for the current period. n-1 ' represents the smoothed heat accumulation of the previous period, α n This represents the decay factor corresponding to the current period. By synchronously smoothing the accumulated heat, the cumulative error fluctuations during the integration process are eliminated.

[0030] Specifically, the first temperature slope sequence is the normalized temperature slope sequence output by S11. This sequence, along with the assigned decay factor sequence, is input, and an adaptive exponential smoothing algorithm is used for iterative calculation. The iterative formula for the exponentially weighted moving average is T. n '=α n ·T n +(1-α n )·T n-1 ', where T n T represents the original temperature slope of the current cycle. n-1 ' is the smoothed value of the previous period, α n This represents the decay factor corresponding to the current period. Since the decay factor for each period is dynamically adjusted based on the period-by-period change, this algorithm can effectively filter out transient thermal disturbances while maintaining the continuity of the degradation trend, and output a second temperature slope sequence.

[0031] For example, the smoothing value of the previous period is 2.4, the original temperature slope of the current period is 3.1, and the decay factor is 0.2. Substituting these values ​​into the iterative calculation, the current smoothing value is T' = 0.2 × 3.1 + 0.8 × 2.4 = 2.54. After iterating through each period, the random fluctuation amplitude of the entire sequence is significantly reduced, and a smooth second temperature slope sequence is output.

[0032] It should be noted that, to quantify the degradation characteristics of the device, after obtaining the smoothed temperature slope sequence, the temperature slope sequence and the corresponding heat accumulation sequence are extracted, and a partial derivative estimation algorithm is used to calculate the rate of change of the temperature slope relative to the heat accumulation. Partial derivative estimation is performed within the main trajectory region identified by the partitioning algorithm, using only (Q,T') data point pairs falling within the main trajectory region, excluding scattered points marked as low-density grids by the partitioning algorithm. The partial derivative is calculated by taking the ratio of adjacent sampling points within the main trajectory region, i.e., r0. n=(T n+1 '-T n ') / (Q n+1 '-Q n The value of T' and Q' is given by the equation '), where T' and Q' represent the temperature slope falling within the main trajectory region and the accumulated heat after smoothing with the attenuation factor, respectively. This rate of change reflects the evolution trend of the temperature slope under unit heat accumulation and is directly used as an estimate of the degradation rate. It should be noted that the dimension of the degradation rate estimate is J. -1 This reflects the evolution trend of the normalized temperature slope under unit heat accumulation. The degradation rate estimate is calculated under a normalized system, and all subsequent thresholds (safety threshold, degradation rate threshold, and health status benchmark value) are calibrated under the same normalized dimension to maintain numerical consistency throughout the entire chain. The degradation rate estimate is compared with the preset health status benchmark value, and the degree of degradation acceleration is evaluated through proportional mapping. The health status benchmark value is determined by selecting the first 100 hours of the initial operation phase of devices from the same batch as the benchmark window. Degradation rate estimates are continuously collected under standard full-load conditions, and the arithmetic mean within this window is calculated as the benchmark value. The selection of this window length is based on 1000 hours of continuous monitoring of 20 devices from the same batch, statistical analysis of the temporal stability of the degradation rate estimate, and the standard deviation of the degradation rate tends to stabilize at the 100-hour node (the standard deviation change within the 10-hour window is less than 5%), and the deviation between the mean degradation rate within 100 hours and the mean within 500 hours is less than 3%, indicating that the 100-hour window can effectively represent the initial health status of the device. When the estimated real-time degradation rate reaches 1.5 times the baseline value, it indicates that the degradation rate has significantly deviated from the normal aging level. The 1.5 times setting is based on the ratio of the 99th percentile of the degradation rate distribution of the same batch of devices in the 500 hours before entering the accelerated degradation stage in accelerated life testing to the mean of the baseline window. This ensures that natural fluctuations in the degradation rate within the normal aging range will not trigger misjudgments, while also promptly detecting accelerated degradation trends.

[0033] The accelerated degradation stage is defined and explained in step S14, and will not be repeated here.

[0034] For example, the accumulated heat Q is the sum of the integral values ​​of each 10ms window in S11, approximately 0.608J per window. After accumulating approximately 164 windows, Q reaches 100, and after accumulating approximately 328 windows, Q reaches 200. During this process, the temperature slope sequence corresponding to the point increases from 2.54 to 2.94, with a change rate r = (2.94 - 2.54) / (200 - 100) = 0.004, i.e., the degradation rate estimate is 0.004. When this estimate reaches 1.5 times the preset benchmark value, the device is determined to have entered the accelerated degradation stage, and the corresponding acceleration coefficient is output. It should be noted that this acceleration coefficient is used as a state marker within stage S12, providing a preliminary qualitative judgment of degradation, while the safety threshold judgment in stage S14 is the final quantitative warning trigger criterion. The two constitute a progressive judgment chain: first, the acceleration trend is initially identified by comparing the benchmark value, and then the warning trigger timing is finally confirmed by the safety threshold judgment.

[0035] It should be noted that S12 continuously outputs degradation rate estimates with a period of 100μs, and the sampling period of the degradation trajectory sequence of S13 is 10s. Taking the system startup time as the starting reference, the sliding window is divided according to the whole 10-second boundary. The median of all degradation rate estimates within each 10s window is taken as the trajectory node value of that window. The downsampled nodes form the degradation trajectory sequence in chronological order.

[0036] In step S13, the estimated degradation rate values ​​are arranged in chronological order to obtain a degradation trajectory sequence. Logarithmic regression is then performed on the degradation trajectory sequence to obtain the predicted degradation trend and the trajectory change slope, including: Mark the sampling time corresponding to the degradation rate estimate to obtain the rate estimation node, and concatenate the rate estimation node in time order to obtain the initial degradation trajectory sequence; The initial degraded trajectory sequence is subjected to time discontinuity detection. If a time discontinuity exists, a linear interpolation algorithm is used to complete the initial degraded trajectory sequence to obtain the degraded trajectory sequence. The degradation trajectory sequence is subjected to least squares logarithmic regression to obtain the predicted degradation trend value and the trajectory change slope.

[0037] Specifically, the input consists of a smoothed estimate of the degradation rate obtained earlier and the sampling time corresponding to that degradation rate recorded by the system's real-time clock. These two values ​​are encapsulated and bound together, outputting a rate estimation node with a time label. The system's real-time clock has millisecond-level precision, and each sampling time is recorded as an absolute timestamp, containing year, month, day, hour, minute, second, and millisecond fields, ensuring that the time order of each node can be uniquely determined over a long period. Rate estimation nodes generated within multiple consecutive sampling periods are arranged in chronological order, and a time series concatenation algorithm is used to connect adjacent nodes end-to-end, outputting an initial degradation trajectory sequence. The time series concatenation algorithm sorts and links nodes based on their sampling time fields, ensuring that the time labels of each node are strictly monotonically increasing. Each rate estimation node contains two fields: the degradation rate value and the sampling time. After concatenation, a scatter plot is formed with time as the horizontal axis and degradation rate as the vertical axis, where each discrete point on the horizontal axis corresponds to a degradation rate observation.

[0038] For example, within a certain sampling period, the obtained smoothed estimate of the degradation rate is 0.004, and the corresponding sampling time is marked as a whole second of that period. These two values ​​are encapsulated to form an independent rate estimation node. Multiple consecutive rate estimation nodes are arranged in chronological order to form an initial degradation trajectory sequence.

[0039] It should be noted that during long-term operation of the device, some sampling period data may be lost due to reasons such as brief interruptions in the communication link, memory write delays, or system resets. Therefore, it is necessary to detect whether there is data loss in the initial degradation trajectory sequence. A time interval comparison algorithm is used to traverse all adjacent node pairs and calculate the time difference between each adjacent node pair. When the time difference is greater than a preset sampling period threshold, a time discontinuity is determined to exist between the adjacent node pairs. The preset sampling period threshold is set to 1.5 times the standard sampling period, which is 10s, and the threshold is 15s. The basis for setting this threshold to 1.5 times is that system clock jitter usually does not exceed ±2s, interrupt processing delay does not exceed ±3s, and the total maximum deviation is 5s. 1.5 times, or 15s, can cover normal delays while effectively identifying real data loss. For the detected time fault, two rate estimation nodes before and after the fault are extracted. A linear interpolation algorithm is used to generate virtual nodes uniformly in the fault interval at a step size of 10s. The degradation rate value of the virtual node is determined by the weighted average of the degradation rate values ​​of the two real nodes before and after the fault and their respective time distances, until the entire fault interval is filled, and the completed degradation trajectory sequence is output.

[0040] For example, with a standard sampling period of 10 seconds, if the timestamps of two adjacent nodes are 10:05:20 and 10:05:50 respectively, and the time difference of 30 seconds is greater than the preset sampling period threshold of 15 seconds, a time discontinuity is determined to exist. The degradation rates before and after the discontinuity are 0.004 and 0.007 respectively. Through linear interpolation, two virtual nodes with values ​​of 0.005 and 0.006 are generated at the missing 30-second and 40-second intervals. After completion, a continuous degradation trajectory sequence is obtained.

[0041] Specifically, the input is the completed degradation trajectory sequence. A least-squares logarithmic regression algorithm is used to fit the time labels and degradation rate nodes in the sequence, outputting a baseline logarithmic fitting curve and the corresponding trajectory slope. The logarithmic regression model is y = a·ln(t) + b, where t is time, y is the degradation rate, a is the trajectory slope, and b is the intercept. Parameters a and b are determined by minimizing the sum of squared residuals between the actual degradation rate values ​​and the model prediction values ​​at each node. Least-squares logarithmic regression can be transformed into a standard linear form on the MCU through variable substitution. Letting t' = ln(t), the model becomes y = a·t' + b. a and b can be directly calculated using the cumulative sum formula without iteration. The trajectory slope a (logarithmic coordinates) characterizes the trend of degradation rate change on the logarithmic time coordinates. Based on this, a time extrapolation function y = a·ln(t) + b is constructed according to the logarithmic regression model. The future target time is input into this function, and the predicted degradation trend value is output.

[0042] It should be noted that under long-term operating stress, the device may experience unrealistic rate mutations due to transient thermal shocks, necessitating the removal of outliers from the initial fitting. The initial fitting residual sequence obtained from least squares logarithmic regression is extracted, with each residual value representing the difference between the actual degradation rate of the corresponding node and the predicted value from the logarithmic fitting curve. A sliding window statistical algorithm is used to extract local anomalies, with a window width of 21 nodes. This width is chosen to consider both statistical stability and local sensitivity; 21 nodes correspond to a 210s observation window, sufficient to establish reliable local statistical characteristics without smoothing out degradation feature differences between adjacent regions due to an excessively large window. For each window, the arithmetic mean and standard deviation of the residuals within the window are calculated, constructing a local normal distribution model centered on that window. The current node residual is compared to a threshold of three times the window's standard deviation. If the current node residual exceeds this threshold, it is considered noise interference and is removed. The preset residual threshold is 3 times the local standard deviation. This threshold is set based on the Laida criterion, which states that under a normal distribution, the probability of data points falling within ±3σ is 99.7%. Values ​​exceeding this range are considered outliers, with the false positive rate controlled below 0.3%. To verify the rationality of the removal operation, the kurtosis and skewness statistics of the residual sequence are calculated before and after removal, confirming that the residual distribution after removal is closer to a normal distribution. After removing all outliers, a fluctuating removal sequence is obtained. A second least squares fit is performed on the fluctuating removal sequence, outputting the corrected degradation trend prediction value and trajectory change slope. The updated logarithmic regression model parameters a and b are used for safety threshold determination and extrapolation prediction in stage S14. That is, a time extrapolation function is constructed using the corrected a and b, outputting the updated degradation trend prediction value for safety threshold comparison in stage S14.

[0043] For example, for a degradation trajectory sequence containing 100 consecutive nodes, the fitting residual for each node is calculated using least squares logarithmic regression. If the residual value at the 45th time node exceeds three times the local standard deviation, that node is removed from the original sequence, leaving the remaining 99 nodes to form a fluctuation elimination sequence. After a second fitting of the fluctuation elimination sequence, the direction of degradation trajectory change is determined to be positively increasing, with a trajectory slope of 0.003 (logarithmic coordinates). This indicates that the device degradation rate exhibits a linear upward trend on the logarithmic time coordinate and a gradually slowing growth process on the absolute time coordinate, consistent with the logarithmic evolution law of gallium nitride transistor on-resistance degradation. After calculating 50 cycles forward using the logarithmic trend extrapolation function, t=150, ln(150)≈5.011. Substituting into y=0.003·ln(t)+0.001, the predicted value is 0.003×5.011+0.001≈0.016. The predicted value of the degradation trend increases from the current 0.013 to about 0.016, reflecting the physical characteristic that the growth rate of degradation under the logarithmic trend gradually slows down with the increase of running time.

[0044] In step S14, if the predicted degradation trend value exceeds a preset safety threshold, the degradation rate deviation is determined based on the trajectory change slope, including: The predicted degradation trend value is compared with the preset safety threshold. If the predicted degradation trend value is greater than the preset safety threshold, the degradation rate of the current sampling period in the degradation trajectory sequence is extracted; The degradation rate is corrected based on the slope of the trajectory change to obtain the current degradation rate; The difference between the current degradation rate and the preset degradation rate threshold is calculated to obtain the degradation rate deviation.

[0045] It should be noted that the preset safety threshold is set based on the statistical results of accelerated life tests conducted on the same batch of gallium nitride transistors under normal operating conditions. The accelerated life test was conducted at an ambient temperature of 150°C. Extrapolation to normal operating conditions was performed using the Arrhenius model. The exemplary activation energy Ea was taken as the median of this activation energy range, 0.45 eV. This value corresponds to the typical failure mechanism activation energy range of 0.4 to 0.5 eV for the on-resistance degradation of gallium nitride devices. During extrapolation, the predicted degradation trend at 150°C was calculated using the Arrhenius acceleration factor AF = exp[(Ea / k) × (1 / T)]. use -1 / T stress Converted to normal operating conditions, where k is the Boltzmann constant, T use The normal operating junction temperature is 358K (85℃), T stress To accelerate the stress junction temperature to 423K (150℃), with an acceleration factor AF≈9.4, the distribution of predicted degradation trends when devices enter the accelerated degradation stage was statistically analyzed. The 10th quantile of this distribution was taken as the safety threshold, ensuring that 90% of devices entering the accelerated degradation stage could be detected, reducing the risk of false alarms. For controlling the false alarm rate of normal devices, the distribution of predicted degradation trends of the same batch of devices in the first 500 hours of the normal aging stage was also analyzed to verify that the selected safety threshold was higher than the 95th quantile of this distribution, ensuring that normal devices would not trigger false alarms at a 95% confidence level. The specific accelerated life test conditions were as follows: 20 gallium nitride transistors of the same batch were placed in an ambient temperature of 150℃ and continuously operated at a switching frequency of 100kHz. The degradation trend prediction value of each device was recorded as a curve changing over time. A 20% increase in on-resistance from the initial value was used as the judgment node for entering the accelerated degradation stage, and the degradation trend prediction value corresponding to this node was extracted as a statistical sample.

[0046] For example, the extrapolated predicted value is 0.016, and the preset safety threshold is 0.008. If the extrapolated predicted value exceeds the safety threshold, it is determined that the device is at risk of entering an accelerated degradation state. At this point, the degradation rate corresponding to the current sampling period is extracted from the degradation trajectory sequence, and the extracted value is 0.008. It should be noted that the extrapolated predicted value of 0.016 in S13 is the extrapolation result for the previous 50 periods, and the degradation rate of 0.008 in the current sampling period is the actual recorded value at the corresponding moment on the degradation trajectory sequence. The difference between the extrapolated predicted value and the actual recorded value reflects the normal discrepancy between the extrapolation model and the measured value. The measured value is used as the starting point for calculating the deviation, and the extrapolated predicted value is only used for safety threshold comparison.

[0047] Specifically, the trajectory slope reflects the long-term trend of the degradation rate, and the current single-point degradation rate is corrected using the trajectory slope. The correction method involves smoothing the current single-point degradation rate using the trajectory slope, i.e., vcorr = β·vcur + (1-β)·(vcur + k×Nd), where k is the trajectory slope, Nd is the number of trajectory sampling periods from the current time to the initial evaluation time, the initial evaluation time is the time corresponding to the first node of the degradation trajectory sequence, and β is the smoothing coefficient set to 0.5. This value is determined by scanning accelerated life test data in a step size of 0.1 within the range of 0.1 to 0.9, selecting the β value that minimizes the root mean square error between the deviation and the actual remaining lifetime. Test results show that the root mean square error of β changes smoothly within the range of 0.4 to 0.6, and the median value of 0.5 is taken as the general setting. This correction integrates instantaneous observation and long-term trend information by taking a weighted average of the current measured value and the trend extrapolation value, avoiding the influence of single-point fluctuations on the deviation calculation. The preset degradation rate threshold is set based on the statistical mean of degradation rate of devices in the same batch during the normal aging stage, μ=0.007 plus 1 standard deviation σ=0.002. The statistical sample is the degradation rate data of 20 devices in the same batch during the first 500 hours of the normal aging stage. Therefore, the threshold = μ + σ = 0.009, which represents the theoretical upper limit of normal degradation rate of the device in the current life cycle stage.

[0048] For example, if the trajectory change slope is 0.0001 per cycle, the current cycle is 80 cycles from the initial evaluation cycle, and the smoothing coefficient β is 0.5, then: v corr =0.5×0.008+0.5×(0.008+0.0001×80)=0.5×0.008+0.5×0.016=0.012.

[0049] The preset degradation rate threshold is 0.009, and the calculated difference yields a degradation rate deviation of 0.003.

[0050] In one implementation, the remaining safe operating time is calculated based on a preset on-resistance limit value, according to the degradation rate deviation and the trajectory change slope, including: The degradation rate deviation and the trajectory change slope are weighted and fused to obtain the comprehensive degradation rate; Calculate the resistance difference between the current on-resistance and the preset on-resistance limit value; The remaining safe operating time is obtained by extrapolating the time axis based on the resistance difference and the overall degradation rate.

[0051] It should be noted that the comprehensive degradation rate comprehensively considers both the immediate degradation deviation reflected by the deviation amount and the long-term degradation trend reflected by the trajectory change slope. Before weighted fusion, the deviation amount and trajectory change slope need to be normalized to unify their dimensions and numerical ranges. The deviation amount is normalized using max-min normalization, mapping the original deviation amount to the range of 0 to 200. The mapping benchmark is the maximum deviation amount observed in accelerated life testing of devices in the same batch; in this example, the maximum deviation amount is 0.005. The trajectory change slope is also normalized to the range of 0 to 200, with the upper limit of normalization set at 0.005 / cycle, representing the maximum trajectory change slope value observed in accelerated life testing. The weight allocation for weighted fusion is based on the correlation between the deviation amount and slope and the actual failure time of the device. Using accelerated life testing data from devices in the same batch, the Pearson correlation coefficients between the deviation amount and slope and the failure time are calculated separately. The squared and normalized Pearson correlation coefficients are then used as weights, w d =0.4624 / (0.4624+0.6724)≈0.4, w k =0.6724 / (0.4624+0.6724)≈0.6. The deviation weight is set to 0.4, and the trajectory change slope weight is set to 0.6. This weighting is based on the fact that in accelerated life testing, the correlation coefficient between the trajectory change slope and device failure time is 0.82, which is higher than the correlation coefficient between deviation and failure time (0.68), indicating that the long-term trend has a stronger predictive ability for the degradation process. The comprehensive degradation rate is calculated as V. comp =w d ×D norm +w k ×K norm D norm and K norm These are the normalized deviation and slope, w, respectively. d and w k These correspond to the weights. It should be noted that V... comp To normalize the overall score, values ​​range from 0 to 200, maintaining consistent dimensions of the degradation rate under the normalized system throughout. In the example, V... comp =50.4, and this normalized composite score is used to perform extrapolation calculations on the subsequent time axis.

[0052] For example, the deviation normalized value is 120, the trajectory change slope normalized value is 4, and the weighted fusion V comp =0.4×120+0.6×4=50.4, that is, the normalized score of the comprehensive degradation rate is 50.4.

[0053] Specifically, the latest smoothed resistance value in the current smoothed on-resistance sequence of the device is obtained. The preset on-resistance limit represents the maximum allowable resistance boundary that allows the device to maintain normal switching characteristics. The preset on-resistance limit is set based on the maximum on-resistance specification value specified in the device datasheet. Exceeding this value will cause switching losses to cause the junction temperature to exceed the device's maximum rated junction temperature. The resistance difference between the current on-resistance and the limit value is calculated. Using a time-axis linear extrapolation, the resistance difference is also mapped to the normalized system, and the normalized resistance change ΔR is calculated. norm =(R limit -R cur) / R limit The time span required to reach the limit value is obtained by dividing the normalized resistance change by the normalized overall degradation rate and then multiplying by the normalized reference time window. Linear extrapolation is used as a first-order approximation of the degradation trend within the short-term prediction interval, i.e., the remaining safe operating time Trem = ΔR. norm / (V comp / 100)×T base T base T is the normalized reference time interval calibrated for devices in the same batch under standard full-load conditions. base The calibration method involves taking newly manufactured devices from the same batch and conducting a constant-temperature aging test under standard full-load conditions (load current 20A, switching frequency 100kHz, ambient temperature 25℃). The time span for each device to rise from its initial on-resistance to the maximum on-resistance specified in the datasheet is recorded. The median of the time spans for all tested devices is taken as T. base The median calibration time for the 20 devices in the same batch was 15,000 hours, so T is taken as an example. base =15000h. If the resistance difference is less than or equal to zero, the device is considered to have exceeded the safety boundary.

[0054] For example, if the current on-resistance is detected to be 150mΩ, the preset on-resistance limit is 450mΩ, the resistance difference is 300mΩ, and the normalized resistance change ΔR norm =300 / 450≈0.667. Normalized overall degradation rate V comp =50.4, time span T rem=0.667 / (50.4 / 100)×15000≈19851h. Based on the current system operating time, after time offset calculation, the remaining safe operating time of the device is determined to be 19851h.

[0055] If the remaining safe working time is lower than the preset safe time limit, a risk warning is triggered, including: If the remaining safe operating time is lower than the preset safe time lower limit, extract the current thermoelectric stress state characteristics of the device; The thermoelectric stress state characteristics are matched with a preset risk threshold matrix to generate a trigger command; The risk warning is triggered by the trigger command.

[0056] It should be noted that the preset safety time limit is set based on the shortest time required for a user to respond after a safety warning for a consumer-grade GaN charger. For consumer-grade charger applications, the process from warning issuance to user replacement involves three stages: warning notification, user confirmation, and product replacement. The warning notification and user confirmation take approximately 24 hours, and product replacement (purchase of a new charger or after-sales replacement) takes approximately 48 hours, totaling 72 hours. The preset safety time limit is set to 72 hours to ensure that users have sufficient time to receive the warning information and safely replace the charger. When the remaining safe operating time is less than 72 hours, the device is determined to have entered a high-risk degradation zone. The on-chip integrated high-precision temperature sensor, gate leakage current monitoring circuit, and drain-source voltage sampling circuit collect the current junction temperature, gate leakage current, and drain-source on-state voltage drop data as the raw input for state feature extraction. The gate leakage current is collected through a precision sampling resistor connected in series in the gate circuit. The sampling resistor value is 10Ω, and the leakage current range is 0 to 10mA. The drain-source on-state voltage drop is acquired synchronously during the on-resistance acquisition process in stage S11, without the need for additional hardware.

[0057] For example, if the remaining safe operating time is 48 hours, the lower limit of the safe time is 72 hours, and the time difference is negative 24 hours, the device is determined to have entered a high-risk degradation range. Real-time data of gate leakage current and junction temperature within this range are collected as the raw data for state feature extraction.

[0058] Specifically, principal component analysis was performed on the raw data of gate leakage current, junction temperature, and drain-source on-state voltage drop. First, a three-dimensional data matrix was constructed, with each row representing an observation vector at a sampling time. The covariance matrix of this matrix was calculated, with each element representing the covariance between two variables. Eigenvalue decomposition was then performed on the covariance matrix to obtain three eigenvalues ​​and their corresponding eigenvectors. The two principal components with the largest eigenvalues ​​were selected to form the eigenvectors representing the current thermoelectric stress state, thus achieving dimensionality reduction from three-dimensional to two-dimensional data. Based on statistical analysis of thermoelectric stress data from 1000 sampling points each of 20 devices in the same batch under three levels of degradation, the cumulative variance contribution rate of the first two principal components was greater than 95%, preserving most of the degradation-related information in the raw data. The three degradation levels are mild, moderate, and severe. Their sample labels are pre-determined by independent physical criteria, not by inversely determined by clustering results. Mild degradation corresponds to an increase in on-resistance of less than 10% relative to the initial on-resistance and a remaining safe operating time of not less than 500 hours; moderate degradation corresponds to an increase in on-resistance of 10% to 20% relative to the initial on-resistance or a remaining safe operating time of 72 to 500 hours; and severe degradation corresponds to an increase in on-resistance of more than 20% relative to the initial on-resistance or a remaining safe operating time of less than 72 hours. These sample labels are consistent with the settings in S14, which use a 20% increase in on-resistance as the threshold for entering the accelerated degradation stage and 72 hours as the lower limit of the safe operating time. Principal component analysis projects the multidimensional raw data to the principal direction with the largest variance, maximizing the preservation of data diversity while reducing dimensionality. The extracted thermoelectric stress feature vectors are input into a preset risk threshold matrix for pattern matching. The preset risk threshold matrix consists of three risk intervals: the feature vector value interval corresponding to mild degradation, the interval corresponding to moderate degradation, and the interval corresponding to severe degradation. The standard feature vector template for each interval is calculated from the labeled samples of the corresponding level. Specifically, the thermoelectric stress feature vectors are first grouped according to the sample labels of mild, moderate, and severe degradation, and the mean of the feature vectors of each group is calculated as the initial template vector. Then, center correction is performed within each level group. When K-means clustering is used for center correction, it is only used to remove outliers that deviate from the center of the level. The level label of the cluster center is determined by the labeled samples of that group, and the degradation level is not determined by the unlabeled clustering results. For different batches of devices, data can be re-collected and the template vector updated during the factory calibration stage. The matching algorithm uses the Euclidean distance nearest neighbor classifier to calculate the Euclidean distance between the current feature vector and the template vectors of each level, and selects the template corresponding to the level with the smallest distance as the matching result. The example boundary values ​​for the three-level risk range are: 0 to 45 for mild degradation, 45 to 75 for moderate degradation, and greater than 75 for severe degradation. The boundary values ​​are determined by the endpoints of the 95% confidence intervals of the first principal component score distribution of devices in the same batch under each level of degradation. The specific values ​​are then calculated by substituting the current batch calibration data.It should be noted that the aforementioned first principal component value range is used for preliminary risk level partitioning. The final risk level is based on the template vector matching result of the nearest neighbor classifier. When the current feature vector is located in the boundary region near 45 or 75 and the interval partitioning result is inconsistent with the nearest neighbor classification result, the nearest neighbor classification result is used as the final matching result. If the difference in Euclidean distance between the current feature vector and the adjacent two-level template vectors is less than 5% of the smaller distance, the matching result is output according to the higher risk level to avoid underestimating boundary samples. The warning signal is input to the instruction parsing engine. The instruction parsing engine looks up the corresponding control strategy table according to the warning level and outputs the trigger instruction for hardware intervention. For example, mild degradation corresponds to a control strategy of reducing the switching frequency by 20%, moderate degradation corresponds to a 50% reduction, and severe degradation corresponds to immediate shutdown.

[0059] For example, the feature vector extracted after PCA dimensionality reduction falls within the matching range of the moderate degradation interval, generating a level-two risk warning signal. Upon receiving the level-two warning signal, the instruction parsing engine looks up a table and outputs a trigger instruction to reduce the switching frequency to 50kHz. After receiving the instruction, the underlying driver circuit modulates the period register value of the pulse width modulation controller, extending the switching period from 10μs to 20μs, effectively reducing the switching frequency from 100kHz to 50kHz. With the reduced switching frequency, the device's switching losses decrease accordingly. Under full-load conditions where conduction losses dominate, the total power consumption decreases, the junction temperature decreases accordingly, and the degradation rate slows down. This provides users with sufficient time to safely replace the charger, enabling early identification and hardware intervention of degradation risks.

[0060] In summary, this invention discloses a method for fault diagnosis of gallium nitride (GaN) charger circuits. This invention obtains the on-resistance time series and junction temperature sensing data of GaN transistors under full-load conditions, and performs synchronous mapping to obtain the heat accumulation and temperature slope. The temperature slope is then differentially processed periodically to obtain an attenuation factor, and weighted smoothing is applied to obtain an estimated degradation rate. These degradation rate estimates are arranged chronologically to obtain a degradation trajectory sequence, and logarithmic regression is performed to obtain a predicted degradation trend and trajectory slope. When the predicted degradation trend exceeds a preset safety threshold, the degradation rate deviation is determined. Based on a preset on-resistance limit, the remaining safe operating time is calculated, and a risk warning is triggered when the value falls below the preset lower limit of the safe operating time. This method achieves degradation risk assessment and quantification of the remaining safe operating time for GaN transistors, effectively avoiding sudden failures in GaN chargers caused by transistor thermal runaway, and significantly improving the reliability and safety of system operation.

[0061] The second embodiment of the present invention provides a gallium nitride charger circuit fault diagnosis system, including: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps of the gallium nitride charger circuit fault diagnosis method as described above.

[0062] It should be noted that the gallium nitride charger circuit fault diagnosis system provided in this embodiment of the invention is used to execute all the process steps of the gallium nitride charger circuit fault diagnosis method in the above embodiment. The working principle and beneficial effects of the two are one-to-one, so they will not be described again.

[0063] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0064] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A method for diagnosing faults in a gallium nitride charger circuit, characterized in that, include: The on-resistance time series and junction temperature sensing data of gallium nitride transistors under full load conditions are obtained and synchronously mapped to obtain the heat accumulation and temperature slope. The temperature slope is differentially analyzed periodically to obtain the attenuation factor. The attenuation factor is then used to perform weighted smoothing on the temperature slope and the heat accumulation to obtain an estimated degradation rate. Arrange the degradation rate estimates in chronological order to obtain a degradation trajectory sequence. Perform logarithmic regression on the degradation trajectory sequence to obtain the degradation trend prediction and trajectory change slope. If the predicted degradation trend value exceeds the preset safety threshold, the degradation rate deviation is determined based on the trajectory change slope. Based on the degradation rate deviation and the trajectory change slope, the remaining safe working time is calculated based on the preset on-resistance limit value. If the remaining safe working time is lower than the preset safety time lower limit, a risk warning is triggered.

2. The gallium nitride charger circuit fault diagnosis method according to claim 1, characterized in that, The process of acquiring the on-resistance time series and junction temperature sensing data of gallium nitride transistors under full load conditions, and performing synchronous mapping to obtain the heat accumulation and temperature slope, includes: The on-resistance time series is subjected to sliding window mid-range filtering to obtain a smoothed on-resistance series; The smooth on-resistance sequence and the device junction temperature sensing data are timestamped to obtain a synchronization mapping matrix; Based on the synchronization mapping matrix, the smooth on-resistance sequence is integrated over time to obtain the accumulated heat, and the junction temperature sensing data of the device is differentiated over time to obtain the temperature slope.

3. The gallium nitride charger circuit fault diagnosis method according to claim 1, characterized in that, The process of periodically differencing the temperature slope to obtain the attenuation factor includes: Calculate the slope difference of the temperature slope in adjacent switching cycles to obtain the cycle-by-cycle change; The attenuation factor is determined by comparing the periodic change with a preset change threshold.

4. The gallium nitride charger circuit fault diagnosis method according to claim 1, characterized in that, The step of using the attenuation factor to perform weighted smoothing on the temperature slope and the accumulated heat to obtain an estimated degradation rate includes: The temperature slope is obtained by performing an exponentially weighted moving average using the attenuation factor; The rate of change of the temperature slope sequence relative to the accumulated heat is calculated to obtain an estimate of the degradation rate.

5. The gallium nitride charger circuit fault diagnosis method according to claim 1, characterized in that, The step of arranging the degradation rate estimates in chronological order to obtain a degradation trajectory sequence, and performing logarithmic regression on the degradation trajectory sequence to obtain the degradation trend prediction and trajectory change slope includes: Mark the sampling time corresponding to the degradation rate estimate to obtain the rate estimation node, and concatenate the rate estimation node in time order to obtain the initial degradation trajectory sequence; The initial degraded trajectory sequence is subjected to time discontinuity detection. If a time discontinuity exists, a linear interpolation algorithm is used to complete the initial degraded trajectory sequence to obtain the degraded trajectory sequence. The degradation trajectory sequence is subjected to least squares logarithmic regression to obtain the predicted degradation trend value and the trajectory change slope.

6. The gallium nitride charger circuit fault diagnosis method according to claim 1, characterized in that, If the predicted degradation trend value exceeds a preset safety threshold, the degradation rate deviation is determined based on the trajectory change slope, including: The predicted degradation trend value is compared with the preset safety threshold. If the predicted degradation trend value is greater than the preset safety threshold, the degradation rate of the current sampling period in the degradation trajectory sequence is extracted; The degradation rate is corrected based on the slope of the trajectory change to obtain the current degradation rate; The difference between the current degradation rate and the preset degradation rate threshold is calculated to obtain the degradation rate deviation.

7. A gallium nitride charger circuit fault diagnosis method according to claim 1, characterized in that, The step of calculating the remaining safe operating time based on the degradation rate deviation and the trajectory change slope, using a preset on-resistance limit value, includes: The degradation rate deviation and the trajectory change slope are weighted and fused to obtain the comprehensive degradation rate; Calculate the resistance difference between the current on-resistance and the preset on-resistance limit value; The remaining safe operating time is obtained by extrapolating the time axis based on the resistance difference and the overall degradation rate.

8. A gallium nitride charger circuit fault diagnosis method according to claim 1, characterized in that, If the remaining safe working time is lower than a preset safe time lower limit, a risk warning will be triggered, including: If the remaining safe operating time is lower than the preset safe time lower limit, extract the current thermoelectric stress state characteristics of the device; The thermoelectric stress state characteristics are matched with a preset risk threshold matrix to generate a trigger command; The risk warning is triggered by the trigger command.

9. A gallium nitride charger circuit fault diagnosis system, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the gallium nitride charger circuit fault diagnosis method according to any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the steps of the gallium nitride charger circuit fault diagnosis method according to any one of claims 1 to 8.

Images