A method and system for transient voltage suppression of silicon carbide power devices

Abstract

This invention relates to the field of power device protection technology, and discloses a method and system for transient voltage suppression of silicon carbide power devices. The method includes: acquiring four-dimensional degradation indices of silicon carbide MOSFETs in each surge absorption channel and performing temperature normalization; training a support vector regression model based on a degradation history database to generate a comprehensive health score and remaining lifetime prediction value for each channel; calculating the degradation-corrected efficiency parameters of each channel based on measured degradation index values ​​to generate a full-channel energy state snapshot matrix; solving a multi-channel surge energy allocation scheme using linear programming with a lifetime balancing penalty term based on the energy state snapshot matrix and health state vector; calculating the PWM control parameters for each channel and executing surge energy absorption and feedback. This invention achieves accurate monitoring of multi-channel degradation states and lifetime-balanced energy allocation.

Description

Technical Field

[0001] This invention relates to the field of power device protection technology, and more specifically, to a method and system for suppressing transient voltages in silicon carbide power devices. Background Technology

[0002] In high-reliability applications such as military power supplies, multi-channel parallel surge voltage absorption circuits are typically constructed using silicon carbide MOSFETs. These circuits leverage the high breakdown electric field strength, wide bandgap, and extremely low on-resistance of silicon carbide to clamp and suppress transient surge voltages at the input. Each channel allocates a surge energy absorption share according to its initially calibrated on-resistance and withstand voltage parameters, and the absorbed surge energy is fed back to the energy storage capacitor bank via a DC-DC converter bridge or dissipated as heat through a bleed resistor.

[0003] However, in high-reliability scenarios involving long-term operation and infrequent maintenance, the degradation rates of silicon carbide MOSFETs in multi-channel surge absorption circuits vary due to differences in surge energy stress and operating temperature. Existing surge energy distribution methods select channels based on initially calibrated fixed on-resistance values. As the silicon carbide MOSFETs in each channel degrade differently, the deviation between the actual on-resistance and the calibrated value gradually increases. Channels with severe degradation, due to their lower initial calibrated on-resistance, continue to be preferentially selected, further accelerating their degradation and ultimately leading to failure, resulting in an unbalanced degradation pattern. Furthermore, relying solely on a single on-resistance or leakage current indicator cannot distinguish between different degradation mechanisms, such as gate oxide defect accumulation caused by surge electrical stress and package degradation caused by thermal cycling. This leads to significant deviations in remaining lifetime predictions in the later stages of degradation, and irreversible and continuous deterioration of the overall transient voltage suppression capability of the surge absorption circuit after channel failure. Summary of the Invention

[0004] This invention provides a method and system for suppressing transient voltages in silicon carbide power devices, solving the technical problems in related technologies such as energy distribution imbalance, decreased system reliability, and shortened overall lifespan caused by device degradation in surge absorption circuits of multi-channel parallel silicon carbide MOSFETs.

[0005] This invention provides a method for suppressing transient voltages in silicon carbide power devices, applied to a power supply system including a surge absorption circuit for multi-channel parallel silicon carbide MOSFETs, comprising the following steps: During the quiet period after the surge voltage suppression event ends, the multidimensional degradation index of silicon carbide MOS transistors in each surge absorption channel is obtained. After temperature normalization processing of the multidimensional degradation index, the multidimensional degradation fingerprint of each channel is generated and stored in the degradation history database. Based on the aforementioned degradation history database, a degradation prediction model for each channel is trained using a support vector regression model. The comprehensive health score and remaining lifespan prediction value for each channel are calculated to form a channel-level health status vector. The actual transmission efficiency of each channel after degradation correction is calculated based on the measured values ​​of degradation index of each channel, and the energy state snapshot matrix of the whole channel is generated by combining the available absorption capacity of each channel and the receivable energy margin of the energy storage capacitor bank. When the input voltage exceeds the surge trigger threshold, the objective function is to minimize the heat dissipation and introduce a lifetime balancing penalty term based on the remaining lifetime prediction value. The energy absorption value of each channel with degradation perception and lifetime balancing is solved by linear programming. The gate PWM control parameters of the DC-DC converter bridge for each channel are calculated based on the energy absorption value of each channel, and each channel performs surge energy absorption and feedback operations in parallel.

[0006] Furthermore, the multidimensional degradation index includes a four-dimensional degradation index: the off-state leakage current obtained under standard reverse bias voltage, the measured on-resistance obtained under fixed gate voltage, the gate threshold voltage obtained by linear region scanning method, and the gate leakage current obtained under fixed gate bias conditions; wherein, the gate threshold voltage is obtained by applying a preset low drain-source voltage to the drain and source terminals of the silicon carbide MOS transistor to make the device work in the linear region, scanning the gate voltage stepwise from low to high with a preset step size, and simultaneously obtaining the curve of drain current changing with gate voltage, and locating the gate voltage corresponding to the inflection point where the drain current increases sharply from near zero as the gate threshold voltage.

[0007] Furthermore, the temperature normalization process includes: Using the silicon carbide wide bandgap carrier concentration temperature model, based on the bandgap width of silicon carbide and the Boltzmann constant, the off-state leakage current is normalized to the reference temperature according to the exponential function relationship. The exponential term of the exponential function is the bandgap width divided by twice the Boltzmann constant and then multiplied by the difference between the reciprocal of the reference temperature and the reciprocal of the measured temperature. Based on the temperature index coefficient of the on-resistance of the silicon carbide MOSFET, the on-resistance is normalized to the reference temperature according to the power function relationship of the ratio of the reference temperature to the measured temperature; based on the temperature coefficient of the threshold voltage of the silicon carbide MOSFET, the gate threshold voltage is temperature compensated according to the linear relationship between the measured temperature and the reference temperature. The normalized four-dimensional degradation index, together with the cumulative absorbed surge energy and cumulative switching count of each channel, are combined to generate a multi-dimensional degradation fingerprint sampling record for each channel. The temperature index coefficient is obtained by linearly fitting the logarithmic relationship between temperature and on-resistance by obtaining the on-resistance of each channel silicon carbide MOS transistor at at least three different temperature points during the initial calibration stage.

[0008] Furthermore, the comprehensive health score is calculated as follows: for each dimension of the four-dimensional degradation index, the ratio of the absolute value of the difference between the failure threshold of that dimension and the current normalized value to the absolute value of the difference between the failure threshold of that dimension and the initial value is calculated, which is used as the normalized health component of that dimension; the normalized health components of each dimension are weighted and summed according to preset weight coefficients to obtain the comprehensive health score; the sum of the weight coefficients is one, and the comprehensive health score ranges from zero to one. A value of one indicates that the device is in an initial healthy state, and a value of zero indicates that the device has reached the failure threshold; wherein, the failure thresholds of each dimension degradation index are defined according to the limit parameters in the silicon carbide MOS transistor datasheet, the off-state leakage current failure threshold is the maximum leakage current value specified in the datasheet, the on-resistance failure threshold is 1.5 times the initial value, the threshold voltage failure threshold is the voltage value corresponding to a 20% offset from the initial value, and the gate leakage current failure threshold is the maximum gate leakage current value specified in the datasheet.

[0009] Furthermore, the step of training the degradation prediction model for each channel using a support vector regression model includes: The degradation index values, cumulative surge energy, and cumulative number of switching are standardized to eliminate dimensional differences. For each channel in the degradation history database that has accumulated a preset threshold, the standardized six-dimensional feature vector is used as input and the corresponding comprehensive health score is used as output. The model is trained using a support vector regression model with radial basis function kernel function to generate degradation prediction models for each channel. The remaining lifetime prediction value is generated as follows: the latest multidimensional degradation fingerprint and stress accumulation of each channel are input into the trained degradation prediction model, and forward prediction iteration is performed with a preset future surge energy increment step. In each iteration, the comprehensive health score output by the previous prediction and the accumulated stress are used as the input of the next iteration until the comprehensive health score drops to the failure threshold. The product of the iteration number and the surge energy increment is recorded as the remaining lifetime prediction value. For channels whose cumulative number of records has not reached the preset threshold, the remaining lifetime is estimated using a linear extrapolation method. Once the number of records reaches the preset threshold, the prediction result from the support vector regression model is used instead.

[0010] Furthermore, the actual transmission efficiency of each channel after degradation correction is calculated as follows: The measured normalized on-resistance of each channel's silicon carbide MOSFET is used as the replacement of the initial calibration value. The input power of each channel's surge absorption terminal is subtracted from the conduction loss determined by the measured on-resistance and the switching loss determined by the degraded switching characteristics, and then divided by the input power to obtain the actual transmission efficiency of each channel. The conduction loss is the product of the square of the drain current and the measured normalized on-resistance. The switching loss is obtained by applying a standard switching test signal to each channel's silicon carbide MOSFET during the silent period, acquiring the energy integral value of the voltage-current overlap interval during the turn-on and turn-off processes, and multiplying it by the switching frequency. The available absorption capacity is the product of the difference between the rated current and the current on-resistance of each channel, the input surge voltage, and the preset typical duration of a single surge suppression event. The acceptable energy margin is half the product of the total capacitance of the energy storage capacitor bank and the difference between the square of the maximum allowable voltage and the square of the current voltage.

[0011] Further, the step of solving for the energy absorption value of each channel through linear programming includes: the objective function being: the minimum sum of the excess surge energy minus the sum of the products of the energy absorption value of each channel and the corresponding actual transmission efficiency, plus a lifetime balance penalty term; the lifetime balance penalty term is the sum of the products of the weighting coefficient and the reciprocal of the standardized remaining lifetime prediction value of each channel, multiplied by the energy absorption value of that channel; the standardized remaining lifetime prediction value is obtained by standardizing the remaining lifetime prediction value using a maximum-minimum normalization method; the constraints include: the energy absorption value of each channel is not less than zero and not greater than the smaller of the available absorption capacity of the channel and the energy reserve that the energy storage capacitor bank can receive divided by the actual transmission efficiency of the channel; the sum of the energy absorption values ​​of all channels is not greater than the excess surge energy value; the energy absorption value of each channel is not greater than the product of the excess surge energy value and the maximum allocation ratio coefficient of a single channel.

[0012] Furthermore, it also includes a progressive channel degradation and decommissioning step: when the overall health score of a channel drops to a preset channel degradation threshold, the upper bound of the available absorption capacity of the channel is linearly reduced according to the ratio of the overall health score to the channel degradation threshold, and the reduced upper bound of the available absorption capacity replaces the original available absorption capacity to re-execute the linear programming solution; when the overall health score of a channel drops to a preset channel decommissioning threshold, the gate drive circuit of the silicon carbide MOS transistor of the channel is set to a permanent off state and the channel is removed from the allocation topology, and the surge energy multi-channel allocation scheme is re-solved on the reduced channel topology; at the same time, the posterior parameters of the degradation prediction model of the decommissioned channel are used as the initialization parameters when updating the support vector regression model of the remaining active channels, and the degradation law information of the channels that have fully experienced the degradation process is used to assist the remaining channels in predicting the remaining lifetime; wherein, the channel decommissioning threshold is less than the channel degradation threshold.

[0013] Furthermore, it also includes: during surge suppression execution, continuously acquiring the transmission current and transmission voltage of each active channel at a preset state update cycle, calculating the actual cumulative energy absorption value of each channel, inputting the deviation between the actual cumulative energy absorption value and the target allocation amount into a proportional-integral controller, dynamically correcting the duty cycle of each channel, and ensuring that the deviation between the actual energy absorption value and the target allocation amount of each channel is within an acceptable range; after each degradation monitoring cycle, re-acquiring the multi-dimensional degradation index of each channel and updating the degradation history database, retraining the degradation prediction model to update the channel-level health state vector, and refreshing the actual transmission efficiency parameters and allocation optimization parameters based on the updated channel-level health state vector, so that the surge energy multi-channel allocation scheme is always solved based on the latest degradation state of each channel; wherein, the duty cycle in the gate PWM control parameters of each channel is determined according to the ratio of the target voltage of the energy storage capacitor group of each channel to the input surge voltage, and the target voltage of the energy storage capacitor group is obtained by reverse calculation using the capacitor energy storage formula based on the allocated absorbed energy value, the current voltage of the energy storage capacitor group, the total capacitance value of the energy storage capacitor group, and the actual transmission efficiency of the channel.

[0014] This invention also proposes a transient voltage suppression system for silicon carbide power devices, comprising: The degradation fingerprint acquisition module is used to acquire multidimensional degradation indices of silicon carbide MOS transistors in each surge absorption channel during the silent period after the surge voltage suppression event ends. After temperature normalization processing of the multidimensional degradation indices, multidimensional degradation fingerprints of each channel are generated and stored in the degradation history database. The degradation prediction module is used to train the degradation prediction model of each channel based on the degradation history database using the support vector regression model, calculate the comprehensive health score and remaining life prediction value of each channel, and form a channel-level health status vector. The energy state assessment module is used to calculate the actual transmission efficiency of each channel after degradation correction based on the measured values ​​of degradation index of each channel, and to generate a full-channel energy state snapshot matrix by combining the available absorption capacity of each channel and the receiveable energy margin of the energy storage capacitor bank. The allocation optimization module is used to minimize the heat dissipation of the channel when the input voltage exceeds the surge trigger threshold. It introduces a lifetime balance penalty term based on the remaining lifetime prediction value and solves the energy absorption value of each channel with degradation perception and lifetime balance through linear programming. The control execution module is used to calculate the gate PWM control parameters of the DC-DC converter bridge of each channel based on the energy absorption value of each channel, and drive each channel to perform surge energy absorption and feedback operations in parallel.

[0015] The beneficial effects of this invention are as follows: This invention obtains the four-dimensional degradation index of silicon carbide MOSFETs and normalizes it using a silicon carbide wide-bandgap carrier concentration-temperature model, enabling the differentiation of different degradation mechanisms with a high signal-to-noise ratio. By using a support vector regression model to fit the nonlinear degradation trend of the multi-dimensional degradation index and cumulative stress, accurate calculation of the actual transmission efficiency after degradation correction is achieved. By introducing a lifetime equalization penalty term into the linear programming objective function, channels with longer remaining lifetimes are allocated more energy absorption, while channels with shorter remaining lifetimes are allocated less energy absorption, achieving a balance in the degradation rate of each channel. Through progressive channel degradation and decommissioning, transient voltage suppression performance maintains a smooth transition during channel degradation, solving the problem of unbalanced degradation caused by fixed calibration values ​​in existing technologies. This results in extending the total service life of multi-channel parallel operation and improving the overall withstand voltage margin. Attached Figure Description

[0016] Figure 1 This is a flowchart of the transient voltage suppression method for silicon carbide power devices provided in the embodiments of the present invention; Figure 2 This is a sub-flowchart of step 1 in the transient voltage suppression method for silicon carbide power devices provided in this embodiment of the invention, which obtains the multidimensional degradation fingerprints of silicon carbide MOS transistors in each surge absorption channel and stores them in the degradation history database. Figure 3 This is a sub-flowchart of step 2 in the transient voltage suppression method for silicon carbide power devices provided in this embodiment of the invention, which trains a degradation prediction model based on a degradation history database and generates a comprehensive health score and remaining lifetime prediction value for each channel. Figure 4 This is a sub-flowchart of step 3 in the transient voltage suppression method for silicon carbide power devices provided in this embodiment of the invention, which calculates the efficiency parameters of each channel after degradation correction and generates a full-channel energy state snapshot matrix. Figure 5 This is a sub-flowchart of step 4 in the transient voltage suppression method for silicon carbide power devices provided in this embodiment of the invention, which solves for a multi-channel surge energy distribution scheme with degradation sensing and lifetime balance. Figure 6 This is a sub-flowchart of step 5 in the transient voltage suppression method for silicon carbide power devices provided in this embodiment of the invention, which calculates the PWM control parameters for each channel and performs surge energy absorption and feedback. Figure 7 This is a schematic diagram comparing the normalized degradation indices of each channel provided in the embodiments of the present invention; Figure 8 This is a schematic diagram of the comprehensive health score and remaining life prediction for each channel provided in the embodiments of the present invention; Figure 9 This is a schematic diagram illustrating the composition of the health score components for each channel provided in an embodiment of the present invention; Figure 10 This is a schematic diagram showing the relationship between the actual transmission efficiency and on-resistance of each channel provided in the embodiments of the present invention; Figure 11 This is a schematic diagram of the energy absorption capacity status of each channel provided in the embodiments of the present invention; Figure 12 This is a schematic diagram of the standardized remaining lifetime and surge energy distribution of each channel provided in an embodiment of the present invention; Figure 13 This is a schematic diagram illustrating the proportion of surge energy distribution scheme provided in the embodiments of the present invention; Figure 14 This is a schematic diagram of the PWM control parameters for each channel provided in the embodiments of the present invention. Detailed Implementation

[0017] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, some features described in the examples may be combined in other examples.

[0018] According to an embodiment of this implementation, such as Figures 1-6As shown, a transient voltage suppression method for silicon carbide power devices is provided, applied to a power supply system including a multi-channel parallel silicon carbide MOSFET surge absorption circuit. It should be understood that the power supply system includes the following hardware environment: multiple parallel silicon carbide MOSFET surge absorption channels, each channel including a silicon carbide MOSFET, a gate drive circuit, and a DC-DC converter bridge composed of silicon carbide MOSFETs; a storage capacitor bank and a discharge resistor branch connected to each channel; voltage sensors, current sensors, and temperature sensors disposed in each channel; and a processing unit for performing degradation monitoring, distribution optimization, and output control.

[0019] The specific steps of this implementation method are as follows: Step 1: Obtain the multidimensional degradation fingerprint of each surge absorption channel silicon carbide MOSFET and store it in the degradation history database; During the quiet period following each surge voltage suppression event, four-dimensional degradation metrics were obtained for the silicon carbide MOSFETs in each surge absorption channel, including: the off-state leakage current obtained under standard reverse bias voltage. Measured on-resistance obtained under a fixed gate voltage Gate threshold voltage and gate leakage current ,in Indicates the first One channel, , This represents the total number of channels.

[0020] Step 101: Set each silicon carbide MOSFET to the off state, apply a standard reverse bias voltage, and obtain the off-state leakage current of each channel. .

[0021] Step 102: Set each silicon carbide MOSFET to the ON state, apply a fixed gate voltage, and obtain the measured on-resistance value of each channel. .

[0022] Step 103: Apply a low drain-source voltage to each silicon carbide MOSFET, scan the gate voltage and locate the turn-on transition point to obtain the gate threshold voltage of each channel. .

[0023] Step 104: Under the condition of applying a fixed bias to the gate, obtain the gate leakage current of each channel. .

[0024] Step 105: Using the silicon carbide wide bandgap carrier concentration temperature model, normalize the off-state leakage current and on-resistance to the reference temperature. Obtain the normalized off-state leakage current and normalized on-resistance .

[0025] The temperature normalization of the off-state leakage current is calculated using the following formula: ; in, The bandgap of silicon carbide is 3.26 eV; Here is the Boltzmann constant, which takes the value of ; For the first The measured temperature when obtaining degradation indicators through each channel; This is the preset reference temperature, typically set to 298K; This represents an exponential function.

[0026] The temperature normalization of the on-resistance is calculated using the following formula: ; in, The temperature index coefficient of the on-resistance of a silicon carbide MOSFET is typically between 1.5 and 2.5.

[0027] Step 106: Utilizing the threshold temperature coefficient of silicon carbide MOSFETs Temperature compensation is applied to the threshold voltage to obtain the compensated threshold voltage. Calculate according to the following formula: ; in, This is the temperature coefficient of the threshold voltage of a silicon carbide MOSFET, which is a negative value and typically ranges from -2mV / K to -5mV / K.

[0028] Step 107: Normalize the four-dimensional degradation index , , , Together with the cumulative absorption of surge energy in each channel and cumulative number of switches Combine the data to generate multidimensional degradation fingerprint sampling records for each channel and store them in the degradation history database for each channel.

[0029] Furthermore, the aforementioned gate threshold voltage is obtained using a linear region scanning method. Specifically, a preset low drain-source voltage is applied across the drain and source terminals of the silicon carbide MOSFET (this low drain-source voltage allows the device to operate in the linear region). Then, the gate voltage is scanned progressively from low to high with a preset step size, simultaneously acquiring the curve of drain current versus gate voltage. The gate voltage corresponding to the inflection point where the drain current begins to increase sharply from near zero is identified as the gate threshold voltage. .

[0030] Furthermore, the temperature exponent coefficient used in the aforementioned silicon carbide wide bandgap carrier concentration temperature model... The temperature index can be the nominal value from the datasheet of the silicon carbide MOSFET, or it can be the actual temperature index coefficient of the device obtained through measured fitting at multiple temperature points during the initial calibration phase. When using the measured fitted value, the on-resistance of each channel of the silicon carbide MOSFET is obtained at at least three different temperature points, and a linear fit is performed on the logarithmic relationship between temperature and on-resistance to obtain the temperature index coefficient corresponding to that channel. .

[0031] Furthermore, the aforementioned cumulative absorption of surge energy This refers to the time from the initial start of operation to the current moment, the [number]th [time]. The sum of the surge energy actually absorbed by each channel during all surge suppression events. After each surge suppression event, the actual energy absorbed by that channel in that event is added to the sum. Cumulative number of switches This refers to the time from the initial start of operation to the current moment, the [number]th [time]. The total number of times a silicon carbide MOSFET in each channel is switched on and off.

[0032] Step 2: Train a degradation prediction model based on the degradation history database to generate a comprehensive health score and remaining life prediction value for each channel; For channels whose cumulative number of records in the degradation history database reaches a preset threshold, the degradation index values ​​and cumulative surge energy are normalized according to each dimension. and cumulative number of switches As input features, the overall health score of the device To output the labels, a degradation prediction model for each channel is trained using a support vector regression model.

[0033] Step 201: Preprocess the normalized degradation index values, cumulative surge energy, and cumulative switching counts obtained in Step 1. Because the dimensions and numerical ranges of the degradation indices differ significantly (e.g., the off-state leakage current is on the order of microamperes, the on-resistance is on the order of milliohms, and the cumulative surge energy is on the order of joules), directly inputting them into the support vector regression model would result in features with large numerical ranges dominating the training process. Therefore, the features of each dimension are standardized to eliminate the impact of dimensional differences on the training of the support vector regression model.

[0034] Step 202: Calculate the overall device health score for each channel. As training labels, they are calculated according to the following formula: ; in, Indexed by degradation metric dimensions. These correspond to the off-state leakage current, on-resistance, threshold voltage, and gate leakage current, respectively. For the first The weighting coefficients of the degradation index satisfy the following: ; For the first The first channel The current normalized value of the degradation index; For the first Failure threshold of degradation index; For the first The initial value of the degradation index. When This indicates that the device is in its initial healthy state. A value dropping to 0 indicates that the device has reached the failure threshold.

[0035] Furthermore, all terms in the above formula are dimensionless quantities. , and All three are values ​​of the same degradation index under different states, and they have the same dimension. Therefore, the numerator... and denominator Since they have the same dimensions, dividing them yields the dimensionless normalized health component. Each dimension's health component is then weighted by a coefficient. After weighted summation, the overall health score is calculated. It is still a dimensionless quantity, with a value range of 0 to 1.

[0036] Furthermore, the failure thresholds for degradation indicators across various dimensions. According to the limiting parameters defined in the datasheet of the silicon carbide MOSFET, specifically, the off-state leakage current failure threshold is the maximum leakage current value specified in the datasheet; the on-resistance failure threshold is 1.5 times the initial value; the threshold voltage failure threshold is the voltage value corresponding to a 20% offset from the initial value; and the gate leakage current failure threshold is the maximum gate leakage current value specified in the datasheet.

[0037] Furthermore, the aforementioned weighting coefficients The weighting coefficients are determined based on the relative importance of each degradation metric to the overall performance of the silicon carbide MOSFET. Generally, on-resistance and gate threshold voltage have a significant impact on the device's turn-on characteristics, and their weighting coefficients are set higher; off-state leakage current and gate leakage current have relatively smaller impacts on the device's turn-off characteristics, and their weighting coefficients are set lower. A typical weighting coefficient allocation scheme is as follows: (Off-state leakage current) (On-resistance) (Threshold voltage) (Gate leakage current).

[0038] Step 203: For each channel whose cumulative number of records in the degradation history database reaches a preset threshold, use the feature vector composed of the standardized degradation index value, cumulative surge energy and cumulative number of switching times of each record in the degradation history database as input, and the corresponding comprehensive health score as output, and use the support vector regression model to train and generate the degradation prediction model for each channel.

[0039] The input to the support vector regression model is a six-dimensional feature vector. The standardized feature vector is output as a comprehensive health score. .

[0040] Furthermore, the aforementioned support vector regression model employs a radial basis function kernel function, determining the support vector regression model parameters by minimizing the sum of the prediction error and the regularization term. Specifically, the support vector regression model training process involves solving a convex optimization problem, selecting the samples that contribute the most to the regression function from the training samples as support vectors, determining the weight coefficients corresponding to each support vector, and finally constructing a nonlinear mapping relationship from the input feature vector to the comprehensive health score.

[0041] Furthermore, the aforementioned preset threshold refers to the minimum number of samples required for the cumulative number of records in the degradation history database to ensure the effectiveness of support vector regression model training. This preset threshold is typically set to 30 to 50 records. When the cumulative number of records for a certain channel does not reach this preset threshold, support vector regression model training for that channel is temporarily suspended; instead, a linear extrapolation method is used for preliminary remaining lifetime estimation.

[0042] Step 204: Input the latest multidimensional degradation fingerprint and stress accumulation of each channel into the trained degradation prediction model, with a preset future surge energy increment. The forward prediction is iterated step by step. In each iteration, the comprehensive health score output from the previous prediction and the accumulated stress are used as the input for the next iteration, until the comprehensive health score drops to the failure threshold. The cumulative surge energy steps of the iteration are recorded, and the current comprehensive health score of each channel is generated. and remaining life prediction (Characterized by remaining tolerable surge energy), forming a channel-level health state vector. ,in This represents the total number of channels.

[0043] Furthermore, the surge energy increment in the aforementioned forward prediction iteration This refers to the step value calculated based on the average energy absorbed by a single surge event in the historical operating records of the power system. In other words, it is the predicted remaining lifetime value. The character represents the first The total amount of surge energy that each channel is expected to absorb under its current degraded state.

[0044] Furthermore, the specific process of the above forward prediction iteration is as follows: First, the current time step (i.e., the...) is... Using the multidimensional degradation fingerprint and stress accumulation of each channel as initial inputs, the degradation prediction model outputs the current comprehensive health score. Then increase the accumulated surge energy. get Simultaneously, the cumulative number of switches is increased accordingly based on the average number of switches per surge event. The updated cumulative stress and the degradation index value corresponding to the previously predicted comprehensive health score are used as inputs, and the comprehensive health score for the next time moment is output again through the degradation prediction model. Repeat the above iterative process until the overall health score drops to a preset failure threshold (usually set to 0.1 to 0.2), and record the number of iterations. Then the remaining life prediction value ,in This represents the number of iterations.

[0045] Furthermore, for channels whose accumulated number of records in the degradation history database has not reached a preset threshold, a linear extrapolation is performed using the degradation index change trends of all existing records. Combined with the failure thresholds of degradation indicators across various dimensions, the initial predicted remaining lifespan of the channel is estimated, and the comprehensive health score is directly calculated using the formula in step 202. This initial estimate will be replaced by the prediction result of the support vector regression model once the number of records accumulates to reach the preset threshold.

[0046] Step 3: Calculate the degradation-corrected efficiency parameters of each channel based on the measured values ​​of the degradation indices of each channel, and generate a full-channel energy state snapshot matrix; Step 301: Based on the measured on-resistance values ​​of each channel's silicon carbide MOSFET Replace the initial calibration values ​​and calculate the actual transmission efficiency of each surge absorption channel. Calculate according to the following formula: ; in, For the first Input power of each channel surge absorption terminal; For the first The receiving power of each channel energy storage terminal; The conduction loss is determined by the measured on-resistance and is calculated using the following formula: ,in For the first The drain conduction current of each channel; The switching loss is determined by the degraded switching characteristics. Generate the degradation-corrected efficiency parameters for each channel. .

[0047] Furthermore, the above-mentioned actual transmission efficiency It is a dimensionless quantity. The numerator in the formula... and denominator Both are power values ​​with the same dimension. Dividing them yields the dimensionless actual transmission efficiency value, which ranges from 0 to 1.

[0048] Furthermore, the aforementioned switching losses The calculation refers to the switching transient energy loss obtained based on the turn-on and turn-off times of each channel's silicon carbide MOSFET after degradation. Specifically, during the quiet period after each surge suppression event, a standard switching test signal is applied to each channel's silicon carbide MOSFET to obtain the energy integral value of the voltage-current overlap interval during the turn-on and turn-off processes. This energy is used as the energy loss of a single switch and then multiplied by the switching frequency to obtain the switching power loss.

[0049] Furthermore, the aforementioned input power This refers to the period during a surge suppression event, the first The instantaneous power absorbed by each channel from the input surge voltage source is calculated by multiplying the measured values ​​from the input voltage sensor and current sensor. The power received at the energy storage end... It refers to the first The instantaneous power transmitted from each channel to the energy storage capacitor bank through the DC-DC converter bridge is calculated by multiplying the measured values ​​of the voltage and current sensors at the energy storage capacitor bank end.

[0050] Step 302: Obtain the available absorption capacity of each surge absorption channel. Calculate according to the following formula: ; in, For the first The rated current of each channel; For the first The current on-state current of each channel; This refers to the input surge voltage. This is the preset typical duration of a single surge suppression event. Simultaneously, the current voltage of the corresponding energy storage capacitor bank for each channel is acquired. and available energy margin The acceptable energy margin is calculated using the following formula: ; in, For the first The total capacitance of the energy storage capacitor bank in each channel; This is the maximum allowable voltage for the energy storage capacitor bank.

[0051] Furthermore, the above-mentioned available absorption capacity The character represents the first The maximum energy that each channel can absorb during the duration of a typical surge suppression event in its current operating state. (Formula omitted for brevity) This indicates the current remaining current margin of the channel. This represents the surge voltage amplitude. The three values, representing the typical surge duration, are multiplied together to obtain the energy dimension of the available absorption capacity.

[0052] Step 303: Calculate the actual transmission efficiency of each channel. Available absorption capacity Current voltage of energy storage capacitor bank and available energy margin Combine to generate a full-channel energy state snapshot matrix. .

[0053] Step 4: Based on the full-channel energy state snapshot matrix and channel-level health state vector, solve for a degradation-aware and lifetime-balanced surge energy multi-channel allocation scheme; When the input voltage exceeds the surge trigger threshold, calculate the excess surge energy value at the input. .

[0054] Furthermore, the aforementioned surge trigger threshold is determined based on the rated operating voltage range of the power supply system, typically set to 1.2 to 1.5 times the upper limit of the rated voltage. (Input excess surge energy value) It is obtained by integrating the portion of the input voltage that exceeds the upper limit of the rated voltage over time.

[0055] Furthermore, the aforementioned excess surge energy value The calculation formula is: ; in, This is the starting moment when the input voltage exceeds the surge trigger threshold. The point at which the input voltage drops below the surge trigger threshold. The measured voltage at the input terminal. The upper limit of the rated voltage. The measured current at the input terminal. The time variable is used. This integral is obtained by numerical integration of the sampled data from the voltage and current sensors.

[0056] Step 401: Using minimizing heat dissipation as the objective function, and the actual transmission efficiency after degradation correction for each channel is used as the objective function. Available absorption capacity And the energy storage capacitor bank can receive energy surplus As a constraint, a lifetime equilibrium penalty term is introduced, and the energy absorption value of each channel is solved by linear programming. The objective function is defined by the following formula: ; in, This represents the current number of active channels. To be assigned to the The energy absorbed by each channel is the decision variable to be solved. For the first Actual transmission efficiency after channel degradation correction; For the first Standardized remaining lifetime prediction for each channel; The weighting coefficient for the lifetime balance penalty term is a dimensionless quantity used to adjust the priority between minimizing the discharge amount and lifetime balance. It typically ranges from 0.1 to 1.0.

[0057] Furthermore, to ensure consistency of dimensions in the objective function, the remaining lifetime prediction value is... Standardization is performed to obtain the dimensionless standardized remaining lifetime value. The standardization process uses the maximum-minimum normalization method, and the calculation formula is as follows: ,in and These are the minimum and maximum predicted remaining lifetime values ​​among all currently active channels, respectively. Weighting coefficients. For a dimensionless quantity, such that The dimensions of the first term are the energy released. Both are units of energy, and the two can be directly added. By adjusting... The numerical value controls the relative weight of the lifetime equilibrium penalty term in the objective function.

[0058] The constraints include: ; The input to the linear programming problem described above is the full-channel energy state snapshot matrix. The actual transmission efficiency parameters, capacity constraints, and channel-level health state vectors for each channel are shown in the figure. The remaining lifetime prediction value is output as the energy absorption value allocated to each channel. .

[0059] Furthermore, the solution process for the aforementioned linear programming problem refers to using standard linear programming algorithms such as the simplex method or interior-point method to search for a combination of decision variables that minimizes the objective function within the feasible region that satisfies all constraints. The solver iteratively updates the energy absorption values ​​of each channel based on the gradient direction of the objective function and the constraint boundaries until it converges to the optimal solution.

[0060] Step 402: Based on the linear programming solution, generate a multi-channel surge energy distribution scheme, including the energy absorption value of each channel. And the remaining discharged energy that cannot be absorbed by the energy storage capacitor bank .

[0061] Furthermore, the aforementioned lifespan balance penalty term The meaning refers to the standardized remaining lifetime value of a certain channel. When the value is large, the penalty term coefficient corresponding to this channel A smaller value means the channel will be allocated more absorbed energy during the minimization of the objective function; conversely, a channel with a shorter remaining lifetime will have a smaller penalty term coefficient. Larger surges are allocated less energy to channels with longer remaining lifespans, thus actively balancing the degradation rate of silicon carbide MOSFETs in each channel.

[0062] Furthermore, to prevent a single channel from experiencing instantaneous overheating due to excessive surge energy under extreme conditions, a maximum allocation ratio constraint for a single channel is introduced into the linear programming constraints described above: ; in, This is the maximum allocation ratio coefficient for a single channel, and its value range is... This value is typically set to 0.5 to 0.7. This constraint ensures that even if the remaining lifetime and available capacity of a channel are significantly higher than those of other channels, the energy absorbed in a single allocation will not exceed a certain percentage of the excess surge energy. times.

[0063] Step 5: Calculate the PWM control parameters for each channel based on the surge energy multi-channel distribution scheme and execute surge energy absorption and feedback; Step 501: Allocate absorbed energy values ​​based on each channel Calculate the gate PWM control parameters of the DC-DC converter bridge composed of silicon carbide MOSFETs in each absorption channel. Calculate the duty cycle of each channel. According to the input surge voltage and the target voltage of each channel energy storage capacitor bank Calculation of the voltage ratio between them: ; in, For the first Each channel energy storage capacitor bank receives, distributes, and absorbs energy. The target voltage after that. The switching frequency of each channel. The power is determined based on the switching characteristics of the silicon carbide MOSFET and the transmission power corresponding to the energy absorbed.

[0064] Furthermore, the aforementioned duty cycle It is a dimensionless quantity. The numerator in the formula... and denominator Both are voltages and have the same dimensions. Dividing them yields a dimensionless duty cycle value, ranging from 0 to 1.

[0065] Furthermore, the target voltage of the aforementioned energy storage capacitor bank According to the distribution of absorbed energy and the current voltage of the energy storage capacitor bank The formula is obtained by reverse calculation using the capacitor energy storage formula. ,in For the first The total capacitance of the energy storage capacitor bank in each channel. This represents the actual transmission efficiency of the channel.

[0066] Furthermore, the aforementioned switching frequency Based on the average transmission power corresponding to the allocated absorbed energy The switching characteristics of the silicon carbide MOSFET are determined. The switching frequency needs to meet the power transmission requirements while also considering the constraints of switching losses and electromagnetic interference. Typically, the switching frequency ranges from 20kHz to 100kHz. When the energy absorption is large and the required power transmission is high, a higher switching frequency is chosen to increase the power density; when the energy absorption is small, a lower switching frequency is chosen to reduce switching losses.

[0067] Step 502: Synchronously output the duty cycle of each channel and switching frequency The corresponding gate drive circuits perform surge energy absorption and feedback operations in parallel, transmitting the allocated surge energy to the corresponding energy storage capacitor bank through the DC-DC converter bridge.

[0068] Step 503: For the remaining discharged energy in the surge energy multi-channel distribution scheme The corresponding part drives the bleeder resistor branch to dissipate heat.

[0069] Furthermore, the topology of the aforementioned DC-DC converter bridge can be either a buck topology or a buck-boost topology, depending on the relative magnitudes of the input surge voltage and the voltage of the energy storage capacitor bank. A buck topology is used when the surge voltage is consistently higher than the energy storage capacitor bank voltage; a buck-boost topology is used when the surge voltage may fall below the energy storage capacitor bank voltage during suppression.

[0070] Furthermore, in order to maintain consistency between the actual absorption amount and the target allocation amount of each channel during surge suppression, the transmission current and transmission voltage of each active channel are continuously acquired at a preset state update cycle, and the cumulative value of the actual absorbed energy of each channel is calculated. The actual accumulated energy absorbed With target allocation The deviation input proportional-integral controller dynamically corrects the duty cycle of each channel. When the input voltage drops back to the nominal range, the DC-DC converter bridges of each absorption channel are shut down in sequence, generating a surge termination signal.

[0071] Furthermore, the aforementioned state update period is typically set to 100 microseconds to 1 millisecond. Actual accumulated absorbed energy value. The result is obtained by time integration of the product of the transmission current and transmission voltage of each channel. The proportional-integral controller then calculates based on the deviation. Output duty cycle correction The updated duty cycle is This ensures that the deviation between the actual energy absorbed by each channel at the end of the surge suppression process and the target allocation is within an acceptable range.

[0072] Furthermore, in order to perform progressive degradation of severely degraded channels in unmaintainable scenarios, the following channel degradation and decommissioning steps are included in addition to step 4: When the overall health score of a certain channel Reduced to the preset channel degradation threshold At that time, the upper limit of the available absorption capacity of the channel is linearly reduced according to the comprehensive health score, and the reduced upper limit of the available absorption capacity is... Calculate using the following formula: ; Among them, the channel degradation threshold It is usually set to 0.5 to 0.6.

[0073] Furthermore, in the above formula and Both are dimensionless comprehensive health scores. Dividing the two yields a dimensionless reduction factor, which is then compared with the usable absorbable capacity, which has the dimension of energy. After multiplication, the reduced upper bound of the available absorption capacity It still retains the energy dimension.

[0074] The upper bound of the reduced available absorption capacity. Replace the available absorption capacity in step 4 Then, the linear programming solution from step 4 is executed again to generate the degraded reconfiguration allocation scheme and the corresponding PWM control parameters for each channel. As a result, the load absorbed by the degraded channel is gradually migrated to the healthy channel.

[0075] When the overall health score of a certain channel Reduced to the preset channel retirement threshold (in When this occurs, the gate drive circuit of the silicon carbide MOSFET in that channel is permanently turned off, and the channel is removed from the allocated topology. In the reduced... The surge energy multi-channel allocation scheme from step 4 is re-solved on the channel topology. Simultaneously, the posterior parameters of the degradation prediction model of the decommissioned channel are used as empirical priors for the degradation prediction of the same type of channel and input into the support vector regression model update process of the remaining active channel to update the parameters of the degradation prediction model of the remaining channel and generate a transient voltage suppression scheme after topology reconstruction.

[0076] Furthermore, the aforementioned channel decommissioning threshold The value is typically set to 0.2 to 0.3 to ensure that the channel is proactively decommissioned before the overall health score drops to an extremely low level and approaches the edge of failure, thus avoiding the impact of sudden failure on the overall surge suppression capability.

[0077] Furthermore, the knowledge transfer of the posterior parameters of the aforementioned decommissioning channel degradation prediction model refers to extracting the support vectors and corresponding weight coefficients after the decommissioning channel's support vector regression model has been trained, and using them as initialization parameters for the remaining active channels of the same model and batch during the next support vector regression model update. In other words, when the support vector regression model for the remaining channels is updated and trained, the posterior parameters of the decommissioned channels are used as prior constraints, thereby utilizing the degradation pattern information of channels that have fully experienced the entire degradation process to assist channels still in the middle of degradation in making more accurate predictions of their remaining lifespan.

[0078] Furthermore, in order to form a continuous closed-loop operation of degradation monitoring, allocation optimization, and lifetime balancing, steps 1 and 2 are re-executed after each degradation monitoring cycle to update the channel-level health status vector of each channel. The actual transmission efficiency parameters in step 3 and the allocation optimization parameters in step 4 are refreshed based on the updated channel-level health state vector. In other words, the measured on-resistance, actual transmission efficiency, available absorption capacity, and predicted remaining lifetime of each channel are periodically updated with the degradation monitoring cycle, so that the surge energy multi-channel allocation scheme in step 4 is always solved based on the latest degradation state of each channel.

[0079] Furthermore, the aforementioned degradation monitoring cycle is determined based on the actual operating environment of the power system and the frequency of surge events. In application scenarios with frequent surge events, the degradation monitoring cycle is set to be shorter, typically once every 10 to 50 surge events; in application scenarios with infrequent surge events, the degradation monitoring cycle can be set to a fixed time interval, typically once every 100 to 500 hours.

[0080] According to the embodiments of this implementation, since four-dimensional degradation indices—off-state leakage current, on-resistance, threshold voltage, and gate leakage current—are obtained for each channel of silicon carbide MOS transistors during the silent period after each surge event, and each index is normalized to a reference temperature using a silicon carbide wide bandgap carrier concentration temperature model, the wide bandgap characteristics of silicon carbide material result in extremely low base values ​​for the above indices at room temperature and relatively small interference introduced by temperature fluctuations. Therefore, the four-dimensional degradation indices can distinguish different degradation mechanisms such as gate oxide defect accumulation caused by surge electrical stress (mainly reflected in increased gate leakage current and threshold voltage drift) and package degradation caused by thermal cycling (mainly reflected in increased on-resistance) with a high signal-to-noise ratio, overcoming the problem that relying solely on a single index cannot distinguish different degradation mechanisms, leading to deviations in remaining lifetime prediction.

[0081] Because the support vector regression model uses a four-dimensional degradation index and cumulative stress as inputs to fit nonlinear degradation trends and perform forward prediction iterations, and standardizes the features of each dimension before training the support vector regression model to eliminate the influence of dimensional differences, the support vector regression model can capture the nonlinear degradation acceleration features generated by multi-stress cross-coupling, replacing the static parameter assumptions based on fixed calibration values. This allows the on-resistance and switching loss of each channel to be updated in real time with the degradation state. Therefore, the actual transmission efficiency after degradation correction calculated based on measured values ​​can accurately reflect the current actual performance of each channel.

[0082] Because a lifetime balancing penalty term is introduced into the linear programming objective function of the surge energy multi-channel allocation scheme, and the remaining lifetime prediction value is standardized and the weight coefficients are reasonably set to ensure that the terms of the objective function are consistent in terms of dimensions, the channel with a longer remaining lifetime is allocated more energy absorption in the optimization solution, and the channel with a shorter remaining lifetime is allocated less energy absorption. Therefore, the degradation rate of the silicon carbide MOSFETs in each channel tends to be balanced, which overcomes the problem of unbalanced degradation in the original scheme where severely degraded channels are continuously overused due to low initial calibration on-resistance. This maximizes the total service life and overall withstand voltage margin of multi-channel parallel operation in non-maintainable scenarios.

[0083] Because a gradual channel degradation and decommissioning process is employed, the absorbed load of degraded channels gradually migrates to healthy channels after the overall health score drops to the channel degradation threshold, rather than abruptly transferring all load when a channel suddenly fails. Therefore, transient voltage suppression performance maintains a smooth transition during channel degradation, avoiding a sharp drop in surge absorption capacity caused by sudden channel failure. Simultaneously, the posterior parameters of the degraded channel degradation prediction model are transferred to the knowledge of the remaining active channels. This allows the remaining lifetime prediction model of the remaining channels to utilize the degradation pattern information of channels that have fully undergone the degradation process, accelerating the convergence of the remaining lifetime prediction and thus continuously ensuring the reliability of the surge voltage absorption circuit under long-term maintenance-free conditions.

[0084] like Figures 7-14 As shown, the steps in the above-described method for suppressing transient voltage in silicon carbide power devices are applied to the following application scenarios, with specific examples as follows: In the power system of a certain military unmanned aerial vehicle (UAV), the airborne power module employs an 8-channel parallel silicon carbide MOSFET surge absorption circuit to suppress transient overvoltages generated during the switching of airborne radar and communication equipment. This UAV performs long-endurance reconnaissance missions, with a single flight duration of up to 72 hours, during which maintenance is impossible. The power system's nominal operating voltage is 270V, and the surge trigger threshold is set at 350V. On May 12, 20XX, during its 158th flight mission, the UAV had accumulated 10,560 hours of operation, experiencing 2,847 surge suppression events.

[0085] Step 1: Obtain the multidimensional degradation fingerprint of the silicon carbide MOSFETs in each surge absorption channel; During the quiet period following the 2847th surge suppression event, the system collected degradation parameters for the silicon carbide MOSFETs in the eight surge absorption channels. For each channel, a standard reverse bias voltage of 600V was applied in the off state to obtain the off-state leakage current; a fixed gate voltage of 18V was applied in the on state to obtain the on-resistance using a 100A test current; the threshold voltage was obtained by scanning the gate voltage in 0.1V steps under a low drain-source voltage of 50V; and the gate leakage current was obtained under a 15V gate bias.

[0086] Table 1. Data collected on original degradation indicators for each channel System set reference temperature Silicon carbide bandgap On-resistance temperature index coefficient Threshold voltage temperature coefficient Taking channel 1 as an example, temperature normalization calculation is performed: Normalized off-state leakage current: Normalized on-resistance: Threshold voltage temperature compensation: Table 2 Normalized Degenerate Fingerprint Data for Each Channel The system combines the normalized four-dimensional degradation index with the cumulative absorbed energy and cumulative switching count of each channel to generate a multi-dimensional degradation fingerprint sampling record, which is then stored in the degradation history database. At this point, the cumulative number of records in the degradation history database for each channel has reached 48, meeting the preset threshold for training the support vector regression model.

[0087] Step 2: Train the degradation prediction model and generate a comprehensive health score; The system extracts 48 historical records from the degradation history database of each channel and standardizes the six-dimensional feature vector (normalized off-state leakage current, normalized on-resistance, compensated threshold voltage, gate leakage current, cumulative absorbed energy, and cumulative switching count). Taking channel 1 as an example, the historical range of its off-state leakage current in the degradation history database is 0.62μA to 1.05μA. After standardization using the maximum and minimum values, the current value is standardized to 0.89.

[0088] The system sets failure thresholds for each degradation metric: off-state leakage current failure threshold is 5.0 μA, on-resistance failure threshold is 1.5 times the initial value of 12.0 mΩ (18.0 mΩ), threshold voltage failure threshold is 20% offset from the initial value of 4.20 V (3.36 V), and gate leakage current failure threshold is 500 nA. The weighting coefficients for each metric are set as follows: , , , .

[0089] Taking Channel 1 as an example, the overall health score is calculated. According to the specific implementation method, the formula for calculating the health score is: The initial values ​​are as follows: , , , Substitute the data: Table 3 Calculation Results of Comprehensive Health Score for Each Channel The system trains support vector regression models for each channel separately, using a radial basis function kernel function with parameters... Regularization parameters After training, the latest degradation fingerprint of each channel is input into the degradation prediction model, using the average energy increment absorbed per surge. Forward prediction iterations are performed. Taking channel 1 as an example, the overall health score gradually decreases from 0.618 during the prediction iteration process. At the 87th iteration, the overall health score drops to the failure threshold of 0.15, therefore the remaining lifetime prediction value... .

[0090] Table 4. Predicted Remaining Lifetime for Each Channel The system generates channel-level health status vectors. Step 3: Calculate the efficiency parameters after degradation correction and generate an energy state snapshot matrix; The system calculates the actual transmission efficiency based on the measured normalized on-resistance of each channel. In the previous surge suppression event, the measured drain on-current of each channel was approximately 85A, the peak surge voltage was 412V, and the surge duration was 2.3ms. Taking channel 1 as an example, the conduction loss and actual transmission efficiency are calculated as follows: Conduction loss: ; The switching loss, obtained through degradation-induced switching characteristic testing, is 12.5 mJ per switch, with a switching frequency of 50 kHz. Therefore: ; Input power: ; Actual transmission efficiency: ; Table 5 Actual transmission efficiency after degradation correction for each channel The system acquires the available absorption capacity and energy storage capacitor bank status for each channel. Each channel has a rated current of 150A, a current current of 28A, a preset surge duration of 2.5ms, and a surge voltage of 412V. Taking channel 1 as an example, the available absorption capacity is calculated as follows: ; The total capacitance of the energy storage capacitor banks in all channels is 4700μF, and the maximum allowable voltage is 450V. The current voltage of the energy storage capacitor bank in channel 1 is 285V, with a usable energy margin of: Table 6 Energy state parameters for each channel The system combines the actual transmission efficiency, available absorption capacity, current voltage of the energy storage capacitor bank, and receiveable energy margin to generate a full-channel energy state snapshot matrix. .

[0091] Step 4: Solve for a surge energy distribution scheme that is both degradation-aware and lifetime-balanced; At 15:37 on May 12, 20XX, the airborne radar system switched operating modes, and the input voltage suddenly surged from 270V to 428V, exceeding the surge trigger threshold of 350V. The system calculated the excess surge energy using data sampled by voltage and current sensors. The system standardizes the predicted remaining lifetime values, minimizing the minimum value. maximum value Taking channel 1 as an example: ; Table 7 Standardized Remaining Life Values ​​for Each Channel The system sets the weight coefficient of the lifespan balancing penalty item. Maximum allocation ratio coefficient for a single channel .

[0092] The objective function of the linear programming is: The constraints include: the energy allocated to each channel must be non-negative and not exceed the limits of available absorption capacity and energy storage margin; the total allocated energy must not exceed excess surge energy; and the energy allocated to a single channel must not exceed 60% of the excess surge energy. Since the normalized remaining lifetime value of channel 5 is 0, to avoid division by zero errors, the system sets the lower limit of its normalized remaining lifetime value to 0.05.

[0093] The system uses the simplex method to solve linear programming problems and converges to the optimal solution after 23 iterations.

[0094] Table 8 Surge Energy Distribution Scheme for Each Channel The total energy absorbed is 31.6 kJ, the total energy received by the energy storage capacitor bank is 30.96 kJ, and the remaining energy is discharged. Heat dissipation needs to be achieved through the bleed resistor branch.

[0095] The allocation results show that channel 7 (252.0 kJ), with the longest remaining lifetime, was allocated the largest amount of absorbed energy, 7.2 kJ, while channel 5 (42.0 kJ), with the shortest remaining lifetime, was allocated only 0.5 kJ, demonstrating the effect of the lifetime balance penalty term.

[0096] Step 5: Calculate PWM control parameters and perform surge absorption and feedback; The system calculates the target voltage of the energy storage capacitor bank based on the energy absorbed by each channel. Taking channel 1 as an example, the current voltage of the energy storage capacitor bank is 285V, the total capacitance is 4700μF, the energy absorbed is 3.8kJ, and the actual transmission efficiency is 0.979. Duty cycle: ; Based on average transmission power Based on the switching characteristics of silicon carbide MOSFETs, select the switching frequency. .

[0097] Table 9 PWM control parameters for each channel The system synchronously outputs the duty cycle and switching frequency of each channel to the gate drive circuit, and the eight channels perform surge absorption and feedback operations in parallel. During surge suppression, the system continuously monitors the actual accumulated energy absorbed by each channel with a 200μs state update cycle and dynamically corrects the duty cycle through a proportional-integral controller. After 2.3ms, the input voltage drops back to 278V, and the system sequentially shuts down the DC-DC converter bridge of each channel, generating a surge termination signal.

[0098] The remaining 1.8kJ of energy is dissipated as heat through the discharge resistor branch, which has a resistance of 15Ω and a discharge time of approximately 3.5ms.

[0099] The data flow in the application can be summarized as follows: Throughout the implementation process, the data starts from the initial original degradation index data of each channel (Table 1), and after temperature normalization and compensation processing, normalized degradation fingerprint data is generated (Table 2). Then, based on the degradation history database, a support vector regression model is trained and the comprehensive health score of each channel (Table 3) and the remaining life prediction value (Table 4) are calculated to form a channel-level health status vector.

[0100] Simultaneously, the actual transmission efficiency of each channel after degradation correction is calculated based on the normalized on-resistance (Table 5). Energy state parameters are generated by combining the available absorption capacity of each channel and the state of the energy storage capacitor bank (Table 6), forming a full-channel energy state snapshot matrix. When a surge event is triggered, the system standardizes the remaining lifetime prediction value (Table 7), inputs the channel-level health state vector and the full-channel energy state snapshot matrix into the linear programming solver, generates a degradation-aware and lifetime-balanced surge energy multi-channel allocation scheme (Table 8), and finally converts it into PWM control parameters for each channel (Table 9) and executes surge absorption and feedback operations.

[0101] The data flow process embodies a complete closed-loop logic from degradation monitoring, health assessment, efficiency correction to allocation optimization, ensuring the balanced lifespan and reliable operation of the multi-channel parallel surge absorption circuit under long-term maintenance-free conditions.

[0102] The embodiments of the present invention have been described above. However, the embodiments are not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make more equivalent embodiments under the guidance of the present embodiments, and all of them are within the protection scope of the present embodiments.

Claims

1. A method for suppressing transient voltages in silicon carbide power devices, applied to a power supply system including a surge absorption circuit for multi-channel parallel silicon carbide MOSFETs, characterized in that, Includes the following steps: During the quiet period after the surge voltage suppression event ends, the multidimensional degradation index of silicon carbide MOS transistors in each surge absorption channel is obtained. After temperature normalization processing of the multidimensional degradation index, the multidimensional degradation fingerprint of each channel is generated and stored in the degradation history database. Based on the aforementioned degradation history database, a degradation prediction model for each channel is trained using a support vector regression model. The comprehensive health score and remaining lifespan prediction value for each channel are calculated to form a channel-level health status vector. The actual transmission efficiency of each channel after degradation correction is calculated based on the measured values ​​of degradation index of each channel, and the energy state snapshot matrix of the whole channel is generated by combining the available absorption capacity of each channel and the receivable energy margin of the energy storage capacitor bank. When the input voltage exceeds the surge trigger threshold, the objective function is to minimize the heat dissipation and introduce a lifetime balancing penalty term based on the remaining lifetime prediction value. The energy absorption value of each channel with degradation perception and lifetime balancing is solved by linear programming. The gate PWM control parameters of the DC-DC converter bridge for each channel are calculated based on the energy absorption value of each channel, and each channel performs surge energy absorption and feedback operations in parallel.

2. The transient voltage suppression method for silicon carbide power devices according to claim 1, characterized in that, The multidimensional degradation index includes four-dimensional degradation indexes: the off-state leakage current obtained under standard reverse bias voltage, the measured on-resistance obtained under fixed gate voltage, the gate threshold voltage obtained by linear region scanning method, and the gate leakage current obtained under fixed gate bias conditions; wherein, the gate threshold voltage is obtained by applying a preset low drain-source voltage to the drain and source terminals of the silicon carbide MOS transistor to make the device work in the linear region, scanning the gate voltage stepwise from low to high with a preset step size, and simultaneously obtaining the curve of drain current changing with gate voltage, and locating the gate voltage corresponding to the inflection point where the drain current increases sharply from near zero as the gate threshold voltage.

3. The transient voltage suppression method for silicon carbide power devices according to claim 2, characterized in that, The temperature normalization process includes: Using the silicon carbide wide bandgap carrier concentration temperature model, based on the bandgap width of silicon carbide and the Boltzmann constant, the off-state leakage current is normalized to the reference temperature according to the exponential function relationship. The exponential term of the exponential function is the bandgap width divided by twice the Boltzmann constant and then multiplied by the difference between the reciprocal of the reference temperature and the reciprocal of the measured temperature. Based on the temperature index coefficient of the on-resistance of silicon carbide MOSFETs, the on-resistance is normalized to the reference temperature according to the power function relationship of the ratio of reference temperature to measured temperature. Based on the threshold voltage temperature coefficient of silicon carbide MOSFETs, temperature compensation is performed on the gate threshold voltage according to the linear relationship between the measured temperature and the reference temperature. The normalized four-dimensional degradation index, together with the cumulative absorbed surge energy and cumulative switching count of each channel, are combined to generate a multi-dimensional degradation fingerprint sampling record for each channel. The temperature index coefficient is obtained by linearly fitting the logarithmic relationship between temperature and on-resistance by obtaining the on-resistance of each channel silicon carbide MOS transistor at at least three different temperature points during the initial calibration stage.

4. The transient voltage suppression method for silicon carbide power devices according to claim 1, characterized in that, The comprehensive health score is calculated as follows: for each dimension of the four-dimensional degradation index, the ratio of the absolute value of the difference between the failure threshold of that dimension and the current normalized value to the absolute value of the difference between the failure threshold of that dimension and the initial value is calculated, and this ratio is used as the normalized health component of that dimension. The normalized health components of each dimension are weighted and summed according to preset weight coefficients to obtain a comprehensive health score. The sum of the weighting coefficients is one, and the comprehensive health score ranges from zero to one. A value of one indicates that the device is in an initial healthy state, and a value of zero indicates that the device has reached the failure threshold. Among them, the failure thresholds of each degradation index are defined according to the limit parameters in the silicon carbide MOSFET datasheet. The failure threshold of the off-state leakage current is taken as the maximum leakage current value specified in the datasheet, the failure threshold of the on-resistance is taken as 1.5 times the initial value, the failure threshold of the threshold voltage is taken as the voltage value corresponding to 20% offset from the initial value, and the failure threshold of the gate leakage current is taken as the maximum gate leakage current value specified in the datasheet.

5. The transient voltage suppression method for silicon carbide power devices according to claim 4, characterized in that, The process of training a degradation prediction model for each channel using a support vector regression model includes: Standardize the degradation index values, cumulative surge energy, and cumulative number of switching operations for each dimension to eliminate dimensional differences; For each channel in the degradation history database that has accumulated a preset threshold of records, a degradation prediction model for each channel is generated by training a support vector regression model with radial basis function kernel function, using a standardized six-dimensional feature vector as input and the corresponding comprehensive health score as output. The remaining lifetime prediction value is generated as follows: the latest multidimensional degradation fingerprint and stress accumulation of each channel are input into the trained degradation prediction model, and forward prediction iteration is performed with a preset future surge energy increment step. In each iteration, the comprehensive health score output by the previous prediction and the accumulated stress are used as the input of the next iteration until the comprehensive health score drops to the failure threshold. The product of the iteration number and the surge energy increment is recorded as the remaining lifetime prediction value. For channels whose cumulative number of records has not reached the preset threshold, the remaining lifetime is estimated using a linear extrapolation method. Once the number of records reaches the preset threshold, the prediction result from the support vector regression model is used instead.

6. The transient voltage suppression method for silicon carbide power devices according to claim 1, characterized in that, The actual transmission efficiency of each channel after degradation correction is calculated as follows: the measured normalized on-resistance of the silicon carbide MOS transistor in each channel is replaced with the initial calibration value. The input power of the surge absorption terminal of each channel is reduced by the conduction loss determined by the measured on-resistance and the switching loss determined by the degradation switching characteristics, and then divided by the input power to obtain the actual transmission efficiency of each channel. The conduction loss is the product of the square of the drain current and the measured normalized on-resistance. The switching loss is obtained by applying a standard switching test signal to each channel silicon carbide MOS transistor during the silent period, obtaining the energy integral value of the voltage-current overlap interval during the turn-on and turn-off processes, and multiplying it by the switching frequency. The available absorption capacity is the product of the difference between the rated current of each channel and the current conduction current, the input surge voltage, and the preset typical duration of a single surge suppression event; the receiveable energy margin is half of the product of the total capacitance of the energy storage capacitor bank and the difference between the square of the maximum allowable voltage and the square of the current voltage.

7. The transient voltage suppression method for silicon carbide power devices according to claim 1, characterized in that, The method of solving for the energy absorption value of each channel through linear programming includes: The objective function is: the minimum sum of excess surge energy minus the sum of the products of the energy absorbed by each channel and the corresponding actual transmission efficiency, plus a lifetime balancing penalty term; the lifetime balancing penalty term is the sum of the products of the weighting coefficient, the reciprocal of the standardized remaining lifetime prediction value of each channel, and the energy absorbed by that channel; the standardized remaining lifetime prediction value is obtained by standardizing the remaining lifetime prediction value using the maximum-minimum normalization method; The constraints include: the energy absorption value allocated to each channel is not less than zero and not greater than the smaller of the channel's available absorption capacity and the energy margin that the energy storage capacitor bank can receive, divided by the channel's actual transmission efficiency; the sum of the energy absorption values ​​allocated to all channels is not greater than the excess surge energy value; and the energy absorption value allocated to each channel is not greater than the product of the excess surge energy value and the maximum allocation ratio coefficient for a single channel.

8. The transient voltage suppression method for silicon carbide power devices according to claim 1, characterized in that, It also includes a progressive channel degradation and decommissioning step: when the overall health score of a channel drops to a preset channel degradation threshold, the upper bound of the available absorption capacity of the channel is linearly reduced according to the ratio of the overall health score to the channel degradation threshold, and the reduced upper bound of the available absorption capacity is used to replace the original available absorption capacity to re-execute the linear programming solution. When the overall health score of a channel drops to the preset channel retirement threshold, the gate drive circuit of the silicon carbide MOS transistor in that channel is set to a permanent off state and the channel is removed from the allocation topology. The surge energy multi-channel allocation scheme is then resolved on the reduced channel topology. At the same time, the posterior parameters of the degradation prediction model of the retired channel are used as the initialization parameters when updating the support vector regression model of the remaining active channels. The degradation law information of the channels that have fully experienced the degradation process is used to assist the remaining channels in predicting their remaining lifetime. Wherein, the channel retirement threshold is less than the channel degradation threshold.

9. The transient voltage suppression method for silicon carbide power devices according to claim 1, characterized in that, Also includes: During surge suppression, the transmission current and transmission voltage of each active channel are continuously acquired at a preset state update cycle. The actual cumulative energy absorbed by each channel is calculated. The deviation between the actual cumulative energy absorbed and the target allocation is input into the proportional-integral controller to dynamically correct the duty cycle of each channel, ensuring that the deviation between the actual energy absorbed by each channel and the target allocation is within an acceptable range. After each degradation monitoring cycle ends, the multidimensional degradation indicators of each channel are reacquired and the degradation history database is updated. The degradation prediction model is retrained and the channel-level health state vector is updated. Based on the updated channel-level health state vector, the actual transmission efficiency parameters and allocation optimization parameters are refreshed, so that the surge energy multi-channel allocation scheme is always solved based on the latest degradation state of each channel. The duty cycle in the gate PWM control parameters of each channel is determined according to the ratio of the target voltage of the energy storage capacitor group to the input surge voltage. The target voltage of the energy storage capacitor group is obtained by reverse calculation using the capacitor energy storage formula based on the allocated energy absorption value, the current voltage of the energy storage capacitor group, the total capacitance value of the energy storage capacitor group, and the actual transmission efficiency of the channel.

10. A transient voltage suppression system for a silicon carbide power device, used to perform the steps of a transient voltage suppression method for a silicon carbide power device as described in any one of claims 1-9, characterized in that, include: The degradation fingerprint acquisition module is used to acquire multidimensional degradation indices of silicon carbide MOS transistors in each surge absorption channel during the silent period after the surge voltage suppression event ends. After temperature normalization processing of the multidimensional degradation indices, multidimensional degradation fingerprints of each channel are generated and stored in the degradation history database. The degradation prediction module is used to train the degradation prediction model of each channel based on the degradation history database using the support vector regression model, calculate the comprehensive health score and remaining life prediction value of each channel, and form a channel-level health status vector. The energy state assessment module is used to calculate the actual transmission efficiency of each channel after degradation correction based on the measured values ​​of degradation index of each channel, and to generate a full-channel energy state snapshot matrix by combining the available absorption capacity of each channel and the receiveable energy margin of the energy storage capacitor bank. The allocation optimization module is used to minimize the heat dissipation of the channel when the input voltage exceeds the surge trigger threshold. It introduces a lifetime balance penalty term based on the remaining lifetime prediction value and solves the energy absorption value of each channel with degradation perception and lifetime balance through linear programming. The control execution module is used to calculate the gate PWM control parameters of the DC-DC converter bridge of each channel based on the energy absorption value of each channel, and drive each channel to perform surge energy absorption and feedback operations in parallel.

Images