A system and method for monitoring junction temperature and current of silicon carbide power field-effect transistors.

Abstract

This invention discloses a system and method for monitoring the junction temperature and current of a silicon carbide power MOSFET, comprising silicon carbide power MOSFETs S1 and S2; a main circuit unit connected to silicon carbide power MOSFETs S1 and S2; a driving unit connected to silicon carbide power MOSFETs S1 and S2; a temperature control unit connected to the package of silicon carbide power MOSFET S2; a junction temperature sampling unit in contact with the chip of silicon carbide power MOSFET S2; a current sampling unit connected to the loop of the main circuit unit; a spectral measurement unit disposed on the surface of silicon carbide power MOSFET S2; and a junction temperature and current detection unit connected to the junction temperature sampling unit, the current sampling unit, and the spectral measurement unit, respectively, for calculating the junction temperature and current of the silicon carbide power MOSFET. This invention can be applied online to monitor the operating junction temperature and current of silicon carbide power devices, and has the advantages of high detection frequency and small current and junction temperature errors.

Description

Technical Field

[0001] This invention relates to the field of power electronic device testing technology, and in particular to a system and method for monitoring the junction temperature and current of a silicon carbide power field-effect transistor. Background Technology

[0002] Thanks to their high operating voltage, high operating temperature, and low on-resistance, silicon carbide power MOSFETs have been widely used in power electronic equipment. However, due to the higher current density, thinner chip thickness, and smaller heat dissipation area of ​​silicon carbide-based devices compared to silicon-based devices, they also face more severe failure problems. Therefore, online monitoring of the junction temperature and operating current of silicon carbide power MOSFETs is of great significance.

[0003] In traditional power device operating status monitoring systems, junction temperature and current measurements are performed independently. Traditional methods for measuring junction temperature in power devices mainly include thermistor contact measurement, thermoelectric parameter measurement, infrared thermal imaging, and thermal impedance network modeling. Among these, thermistor contact measurement suffers from large static errors due to the temperature measurement point being far from the power device in high-voltage applications, and its response time is only on the order of seconds. The thermoelectric parameter measurement method relies on temperature-dependent electrical parameters of the device, making it susceptible to electromagnetic interference under the high-frequency switching conditions of silicon carbide power MOSFETs. The infrared thermal imaging method requires expensive and bulky far-infrared spectroscopy imaging equipment and necessitates damaging the power device package to blacken the temperature measurement surface, making it unsuitable for online industrial applications. The thermal impedance network modeling method requires thermally coupled parameters that are difficult to obtain and cannot eliminate the influence of device aging on the measurement. Traditional power device current measurements are primarily performed using sensing elements such as coaxial resistors, current transformers, Hall sensors, and Rogowski coils. Coaxial resistors introduce conduction losses, making it difficult to balance measurement accuracy and power consumption in high-capacity applications. Current transformers and Rogowski coils can only measure alternating current, and current transformers suffer from hysteresis and magnetic saturation, while Rogowski coils are susceptible to interference from ambient magnetic fields. Hall effect sensors exhibit temperature drift, making them unsuitable for the high-temperature operating conditions of silicon carbide power devices. Therefore, traditional methods for detecting the junction temperature and current of power devices are difficult to apply to the high-temperature, high-voltage, and high-power applications of silicon carbide power MOSFETs. There is an urgent need for condition monitoring methods that meet the requirements of high-temperature heat flow isolation, high-voltage electrical isolation, resistance to aging and electromagnetic interference, and easy packaging and integration. Electroluminescence, a light-emitting phenomenon caused by the conduction of the parasitic diode in a silicon carbide power MOSFET, has spectral characteristics coupled with the device's operating conditions, opening up a new avenue for non-invasive online condition monitoring.

[0004] The body diode of a silicon carbide power field-effect transistor (SiCPT) is a bipolar device. When it is turned on, electrons and holes recombine in the lightly doped region of the device, and some of the energy generated is emitted in the form of photons. Both direct interband recombination and deep-level recombination of charge carriers can generate photons, but the energies produced are different, resulting in two main characteristic peaks in the spectrum. The emission intensity of the two characteristic peaks has different sensitivities to the junction temperature and operating current of the device. Therefore, the junction temperature and current can be decoupled and extracted by narrowband filtering of the spectrum.

[0005] Existing device state monitoring methods based on the electroluminescence effect utilize discrete photodiodes and detection circuits to achieve synchronous measurement of junction temperature and current under conditions of continuous conduction or fixed duty cycle of the body diode. However, in real-world operating conditions, silicon carbide power MOSFETs cannot function solely as body diodes; the device may only emit light during the dead time. This reduces the signal-to-noise ratio of the optical signal, requiring the sampling circuit to be designed with higher gain, which in turn implies a higher sampling delay. Summary of the Invention

[0006] The purpose of this invention is to provide a silicon carbide power MOSFET junction temperature and current monitoring system. This invention can be applied online to monitor the operating junction temperature and current of silicon carbide power devices, and has the advantages of high detection frequency and small current and junction temperature errors.

[0007] The technical solution of the present invention: a silicon carbide power field-effect transistor junction temperature and current monitoring system, comprising silicon carbide power field-effect transistors S1 and S2;

[0008] The main circuit unit is connected to silicon carbide power MOSFETs S1 and S2;

[0009] The drive unit, connected to silicon carbide power MOSFETs S1 and S2, is used to provide switching control signals;

[0010] The temperature control unit, connected to the package of the silicon carbide power MOSFET S2, is used to regulate the ambient temperature.

[0011] The junction temperature sampling unit is in contact with the chip of the silicon carbide power field-effect transistor S2 and is used to measure the actual operating junction temperature in the calibration test process.

[0012] The current sampling unit is connected to the loop of the main circuit unit and is used to measure the actual working current in the calibration test stage.

[0013] A spectral measurement unit is disposed on the surface of the silicon carbide power field-effect transistor S2, and is used to convert the complete electroluminescence spectrum of the silicon carbide power field-effect transistor S2 into an electrical signal.

[0014] The junction temperature and current detection unit is connected to the junction temperature sampling unit, the current sampling unit, and the spectral measurement unit, respectively, and is used for calculating the junction temperature and current of silicon carbide power field-effect transistors.

[0015] The aforementioned silicon carbide power MOSFET junction temperature and current monitoring system includes a main circuit unit comprising a DC voltage source Vin, an inductor L, an energy storage capacitor C, and a load resistor R. The positive terminal of the DC voltage source Vin is connected to the drain of the silicon carbide power MOSFET S1, and the negative terminal of the DC voltage source Vin is connected to the source of the silicon carbide power MOSFET S2, one end of the energy storage capacitor C, and one end of the load resistor R. The source of the silicon carbide power MOSFET S1 is connected to the drain of the silicon carbide power MOSFET S2 and one end of the inductor L. The other end of the inductor L is connected to the other end of the energy storage capacitor C and the other end of the load resistor R.

[0016] The aforementioned silicon carbide power MOSFET junction temperature and current monitoring system includes a driving unit comprising two silicon carbide power MOSFET driving circuits and a DC voltage source for powering the driving circuits; one driving circuit is connected to the gate and source of silicon carbide power MOSFET S1, and the other driving circuit is connected to the gate and source of silicon carbide power MOSFET S2.

[0017] The aforementioned silicon carbide power MOSFET junction temperature and current monitoring system includes a spectral measurement unit comprising an entrance slit, a blazed grating, and a CMOS linear image sensor. The spectral measurement unit is fixed to the surface of the silicone layer of the silicon carbide power MOSFET S2 to prevent stray ambient light from entering the device package. The entrance slit is arranged perpendicular to the chip edge of the silicon carbide power MOSFET S2. The electroluminescent beam of the silicon carbide power MOSFET S2 is dispersed by the blazed grating after passing through the entrance slit, and light of different wavelengths converges into different pixels of the CMOS linear image sensor.

[0018] The aforementioned silicon carbide power MOSFET junction temperature and current monitoring system includes a junction temperature detection unit with a built-in FPGA / DSP / microcontroller, which calculates the junction temperature and current.

[0019] The aforementioned method for monitoring the junction temperature and current of a silicon carbide power MOSFET includes the following steps:

[0020] (1) Calibration test: In the calibration test, the electroluminescence intensity vector, the actual working junction temperature and actual working current of the device under various operating conditions are stored to obtain the dataset;

[0021] (2) Model building stage: In the model building stage, the dataset of the calibration test stage is preprocessed, and then the mapping model between the electroluminescence intensity vector and the actual working junction temperature and the actual working current is established based on the partial least squares regression algorithm. The mapping model is trained using the dataset.

[0022] (3) Online monitoring: Based on the trained mapping model and the electroluminescence intensity vector measured by the spectral measurement unit, the working junction temperature and working current of the silicon carbide power field-effect transistor are calculated online.

[0023] The aforementioned method for monitoring the junction temperature and current of a silicon carbide power MOSFET, wherein the dataset is expressed as follows:

[0024]

[0025] In the formula: I ELdata Represents the luminous intensity matrix; I ELij T represents the luminous intensity of the j-th band in the spectrum under the i-th operating condition, where i∈{1,2,3,...,m} and j∈{1,2,3,...,n}; data T represents the junction temperature vector. i I represents the actual junction temperature under the i-th operating condition; data Represents the operating current vector; I i This represents the actual operating current under the i-th operating condition.

[0026] The aforementioned method for monitoring the junction temperature and current of a silicon carbide power MOSFET includes the following preprocessing:

[0027] First, Savitzky-Golay convolution smoothing is applied to the luminescence intensity data to effectively preserve spectral features while removing spectral measurement noise. The convolution smoothing window is set to 2p+1, and the order is q. Then, for the luminescence intensity I of the j-th band in the i-th working condition sample... ELij Using the luminous intensity I of adjacent bands ELi(j-p) I ELi(j-p+1) ... I ELi(j+p-1) I ELi(j+p) polynomial fitting value I ELij' Replace I ELij The fitted polynomial is:

[0028] y = a0 + a1x + a2x 2 +K+a q x q ;

[0029] Where a0, a1, ..., a q These are the fitting coefficients;

[0030] This yields a system of linear equations with q+1 variables, in matrix form:

[0031]

[0032] Where e j-p e j-p+1 ..., e j+p For fitting residuals;

[0033] The above formula can be simplified as follows:

[0034] I ELS_G =P S_G ·A S_G +E S_G ;

[0035] Fitting coefficient matrix A S_G The least squares solution is:

[0036] A S_G =(P S_G T P S_G ) -1 P S_G T I ELS_G ;

[0037] Thus, I is obtained ELij polynomial fitting value I ELij′ ,:

[0038] I ELij = [1 0 0] 1 L 0 q A S_G =a0;

[0039] For each sample, the luminescence intensity of bands p+1 to np is convolved and smoothed to obtain the denoised luminescence intensity matrix I. ELdata' :

[0040]

[0041] Subsequently, m1 groups of samples were randomly selected from the denoised dataset to form the training set, which includes the training set luminescence intensity matrix I. ELtrain Training set working temperature vector T train and the working current vector I of the training set train The expression is:

[0042]

[0043] Similarly, the remaining m-m1 groups of samples were used to form the test set, which includes the test set luminescence intensity matrix I. ELtest Test set working temperature vector T testand test set operating current vector I test :

[0044]

[0045] The aforementioned method for monitoring the junction temperature and current of a silicon carbide power MOSFET includes the following steps for training the mapping model using a dataset:

[0046] The training and test set data are standardized based on the training set sample distribution: First, the mean luminous intensity μ of the training set is calculated. IELj σ, the standard deviation of the luminous intensity of the training set IELj The average working temperature of the training set (μ) T Training set working temperature standard deviation σ T The average operating current of the training set, μ I Training set operating current standard deviation σ I :

[0047]

[0048] The training and test sets were then standardized to obtain the standardized training set luminescence intensity matrix I. ELtrain '、Standardized training set working result vector T' train 'Standardized training set working current vector I' train 'Standardized test set luminescence intensity matrix I' ELtest 'Standardized test set working temperature vector T' test 'and standardized test set operating current vector I test ', its expression is as follows:

[0049]

[0050]

[0051] For the standardized training set, the partial least squares regression algorithm is used to construct the luminescence intensity matrix I of the standardized training set of independent variables. ELtrain 'Working vector T of the standardized training set train The relationship between them: First, calculate the first principal component t1 of the independent variable, its expression is:

[0052]

[0053] Where w1 is the coefficient vector of the first principal component of the independent variable, and λ1 is the matrix I. ELtrain ′、T train ′、T train ′、I ELtrain The largest eigenvalue of ′;

[0054] Subsequently, I was established.ELtrain ′、T train The regression equation between y′ and y1 is:

[0055]

[0056] Where p1 is the regression coefficient vector, r1 is the regression coefficient, and I ELtrain1 ′ and T train1 ′ are the residual matrices of the equation;

[0057] The expression for solving the regression coefficients is:

[0058]

[0059] Regress the equation residuals again to find the principal component t of the i-th independent variable. i Principal component coefficient vector w i , regression coefficient vector p i and regression coefficient r i The expression is:

[0060]

[0061]

[0062]

[0063]

[0064] By selecting the first k principal components to establish a mapping function model, the standardized luminescence intensity vector I of any sample can be obtained. EL 'Regression results with standardized working temperature T' reg 'Working junction temperature function model:'

[0065]

[0066] And to obtain the normalized luminescence intensity vector I of any sample EL 'Regression results with standardized operating current I' reg 'Working current function model between:'

[0067]

[0068] The aforementioned method for monitoring the junction temperature and current of a silicon carbide power MOSFET includes, in the training step, verifying the accuracy of the trained mapping model by standardizing the luminous intensity matrix I of the test set. ELtest Substituting into the working junction temperature function model, we obtain the standardized working junction temperature regression result vector T for the test set. test_reg ', the standardized test set luminescence intensity matrix I ELtestSubstituting into the operating current function model, we obtain the standardized operating current regression result vector I for the test set. test_reg Denormalizing both yields the working junction temperature regression vector T. test_reg and the working current regression result vector I test_reg Its expression is:

[0069]

[0070] Calculate the mean absolute error (MAE) of the working junction temperature of the test set. Ttest and the mean absolute error of the operating current (MAE) Itest :

[0071]

[0072] If the mean absolute error does not meet the requirements, increase the principal component dimension k of the model and recalculate the regression values ​​until a mapping model with a mean absolute error that meets the requirements is selected.

[0073] Compared with existing technologies, this invention is based on the electroluminescence effect of silicon carbide power devices. Through a spectral measurement unit equipped with a spectroscopic dispersion structure and a high-speed CMOS sensor, it obtains high-resolution electroluminescence spectra of the devices and converts them into electrical signals, which are then input to a junction temperature and current monitoring unit for calculation of junction temperature and current. The method of this invention utilizes a training dataset obtained from calibration measurements and combines a partial least squares regression algorithm to establish mapping models between the emission spectra and the device's operating junction temperature and current. This model can be applied online to monitor the operating junction temperature and current of silicon carbide power devices and has the advantages of high detection bandwidth and small errors in current and junction temperature. This invention achieves low-latency detection while ensuring signal strength (i.e., accuracy). Attached Figure Description

[0074] Figure 1 This is a schematic diagram of the system structure of the present invention;

[0075] Figure 2 This is a schematic diagram of the optical path of the spectral measurement unit;

[0076] Figure 3 This is a flowchart illustrating the method of the present invention;

[0077] Figure 4 This is a schematic diagram of the operating signal timing of a silicon carbide device;

[0078] Figure 5 This is a schematic diagram of the partial least squares regression results for the operating junction temperature of a silicon carbide device;

[0079] Figure 6 This is a schematic diagram of the partial least squares regression results for the operating current of silicon carbide devices. Detailed Implementation

[0080] The present invention will be further described below with reference to the accompanying drawings and embodiments, but this should not be construed as limiting the present invention.

[0081] Example 1: A silicon carbide power MOSFET junction temperature and current monitoring system, such as Figure 1 As shown, it includes silicon carbide power field-effect transistors S1 and S2;

[0082] The main circuit unit is connected to silicon carbide power MOSFETs S1 and S2. The main circuit unit includes a DC voltage source Vin, an inductor L, an energy storage capacitor C, and a load resistor R. The positive terminal of the DC voltage source Vin is connected to the drain of the silicon carbide power MOSFET S1, and the negative terminal of the DC voltage source Vin is connected to the source of the silicon carbide power MOSFET S2, one end of the energy storage capacitor C, and one end of the load resistor R. The source of the silicon carbide power MOSFET S1 is connected to the drain of the silicon carbide power MOSFET S2 and one end of the inductor L. The other end of the inductor L is connected to the other end of the energy storage capacitor C and the other end of the load resistor R.

[0083] The driving unit is connected to silicon carbide power field-effect transistors S1 and S2 and is used to provide gate drive signals to control the turn-on and turn-off of the device. The driving unit includes two silicon carbide power field-effect transistor driving circuits and a DC voltage source to power the driving circuits. One driving circuit is connected to the gate and source of silicon carbide power field-effect transistor S1, and the other driving circuit is connected to the gate and source of silicon carbide power field-effect transistor S2.

[0084] The temperature control unit, connected to the package of the silicon carbide power MOSFET S2, is used to regulate the ambient temperature of the device during calibration testing and online operation.

[0085] The junction temperature sampling unit is in contact with the silicon carbide power field-effect transistor S2 chip and is used to measure the actual operating junction temperature in the calibration test process; it should be noted that the junction temperature sampling unit is only used in the calibration test process.

[0086] The current sampling unit is connected to the loop of the main circuit unit and is used to measure the actual working current in the calibration test. It should be noted that the current sampling unit is only used in the calibration test.

[0087] A spectral measurement unit, disposed on the surface of the silicone layer of the silicon carbide power field-effect transistor S2, is used to convert the complete electroluminescence spectrum of the silicon carbide power field-effect transistor S2 into an electrical signal; such as Figure 2As shown, the spectral measurement unit includes an entrance slit, a blazed grating, and a CMOS linear image sensor, used to convert the complete electroluminescence spectrum of the silicon carbide power field-effect transistor into an electrical signal; the optical path structure of the spectral measurement unit is as follows. Figure 3 As shown, the unit is fixed to the surface of the silicone layer of the silicon carbide power device to prevent stray ambient light from entering the device package. The incident slit is arranged perpendicular to the edge of the device chip to allow for maximum electroluminescence intensity. After passing through the slit, the light beam is dispersed by a blazed grating, and light of different wavelengths converges into different pixels of the CMOS linear image sensor. Each pixel of the CMOS linear image sensor converts the accumulated photon signal within a set integration time into a voltage signal, which is then converted from analog to digital to obtain the luminescence intensity vector I of the device's electroluminescence spectrum. EL The vector is composed of the relative luminescence intensities of each band of the electroluminescence spectrum, and the number of vector elements is equal to the number of pixels in the linear image sensor.

[0088] The junction temperature and current detection unit is connected to the junction temperature sampling unit, the current sampling unit, and the spectral measurement unit, respectively, and is used for calculating the junction temperature and current of silicon carbide power field-effect transistors.

[0089] During the calibration test, the electroluminescence intensity vector I under various operating conditions is stored. EL The actual junction temperature T and actual operating current I of the device are used as the basis for model building. These data are preprocessed, and the electroluminescence intensity vector I is established based on the partial least squares regression algorithm. EL A mapping model between the actual junction temperature T and the actual operating current I of the device; and an electroluminescence intensity vector I measured by the established model and the spectral measurement unit during online monitoring. EL The system calculates the operating junction temperature and current of silicon carbide power MOSFETs online. The junction temperature and current monitoring unit incorporates an FPGA / DSP / microcontroller, which enables the establishment of the mapping model and the calculation of junction temperature and current.

[0090] Example 2: A monitoring method based on the silicon carbide power MOSFET junction temperature and current monitoring system in Example 1, such as... Figure 3 As shown, it includes the following steps:

[0091] (1) Calibration test: In the calibration test, the electroluminescence intensity vector, the actual working junction temperature and actual working current of the device under various operating conditions are stored to obtain the dataset;

[0092] Specifically, to establish a model relating the electroluminescence spectrum, operating junction temperature, and current of a silicon carbide power MOSFET under actual operating conditions, calibration tests were performed on normally functioning silicon carbide power devices. Under conditions not exceeding the maximum operating voltage, maximum operating current, and maximum operating junction temperature for safe operation of the silicon carbide power MOSFET, m different operating conditions were set. The ambient temperature of the device was adjusted by a temperature control unit, the operating current was adjusted by changing the resistance of the load resistor R, and the DC voltage source V was adjusted... in The input voltage is used to change the device's turn-off voltage, and the voltage across the load resistor R is kept constant by adjusting the device's switching duty cycle to simulate actual working conditions.

[0093] Under the i-th operating condition, after the system stabilizes, the luminescence intensity vector I of the accumulated electroluminescence spectrum within a set integration time is acquired by the spectral measurement unit. ELi It contains n elements, I ELij This represents the luminescence intensity of the j-th band in the spectrum. Simultaneously, the temperature sampling unit acquires the actual operating junction temperature T of the device. i The current sampling unit collects the operating current I of the device. i The timing sequence of each electrical signal is as follows: Figure 4 As shown, where V g1 V is the drive signal for the silicon carbide power MOSFET S1. g2 The driving signal for silicon carbide power MOSFET S2, i D2 This represents the current flowing through the body diode D2 of the light-emitting silicon carbide power field-effect transistor S2. In each switching cycle T... S Within, the body diode only operates during two dead time periods T. d When the light is emitted, i is turned on. D2 Equal to the current I flowing through inductor L i Therefore, current I can be used. i The actual operating current characterizing the device when it emits light.

[0094] Luminous intensity vector I under each working condition ELij Device operating junction temperature T i and device operating current I i The data samples are stored in the junction temperature and current monitoring unit, forming one data sample. All data samples from the m operating conditions constitute the dataset, including the luminous intensity matrix I. ELdata Device junction temperature vector T data and device operating current vector I data Therefore, the dataset is expressed as follows:

[0095]

[0096] In the formula: I ELdata Represents the luminous intensity matrix; IELij T represents the luminous intensity of the j-th band in the spectrum under the i-th operating condition, where i∈{1,2,3,...,m} and j∈{1,2,3,...,n}; data T represents the junction temperature vector. i I represents the actual junction temperature under the i-th operating condition; i This represents the actual operating current under the i-th operating condition.

[0097] (2) Model building stage: In the model building stage, the dataset of the calibration test stage is preprocessed, and then the mapping model between the electroluminescence intensity vector and the actual working junction temperature and the actual working current is established based on the partial least squares regression algorithm. The mapping model is trained using the dataset.

[0098] The preprocessing is as follows:

[0099] First, Savitzky-Golay convolution smoothing is applied to the luminescence intensity data to effectively preserve spectral features while removing spectral measurement noise. The convolution smoothing window is set to 2p+1, and the order is q. Then, for the luminescence intensity I of the j-th band in the i-th working condition sample... ELij Using the luminous intensity I of adjacent bands ELi(j-p) I ELi(j-p+1) ... I ELi(j+p-1) I ELi(j+p) polynomial fitting value I ELij' Replace I ELij The fitted polynomial is:

[0100] y = a0 + a1x + a2x 2 +K+a q x q ;

[0101] Where a0, a1, ..., a q These are the fitting coefficients;

[0102] This yields a system of linear equations with q+1 variables, in matrix form:

[0103]

[0104] Where e j-p e j-p+1 ..., e j+p For fitting residuals;

[0105] The above formula can be simplified as follows:

[0106] I ELS_G =P S_G ·A S_G +E S_G ;

[0107] Fitting coefficient matrix A S_G The least squares solution is:

[0108] A S_G =(P S_G T P S_G ) -1 P S_G T I ELS_G ;

[0109] Thus, I is obtained ELij polynomial fitting value I ELij' ,:

[0110] I ELij = [1 0 0] 1 L 0 q A S_G =a0;

[0111] For each sample, the luminescence intensity of bands p+1 to np is convolved and smoothed to obtain the denoised luminescence intensity matrix I. ELdata' :

[0112]

[0113] Subsequently, m1 groups of samples (accounting for 75% of the total samples) were randomly selected from the denoised dataset to form the training set, which includes the training set luminescence intensity matrix I. ELtrain Training set working temperature vector T train and the working current vector I of the training set train The expression is:

[0114]

[0115] Similarly, the remaining m-m1 groups of samples were used to form the test set, which includes the test set luminescence intensity matrix I. ELtest Test set working temperature vector T test and test set operating current vector I test :

[0116]

[0117] To ensure consistency in the magnitude range of spectral features across different bands and improve model accuracy, the training and test set data are standardized based on the training set sample distribution: First, the mean luminescence intensity μ of the training set is calculated. IELj σ, the standard deviation of the luminous intensity of the training set IELj The average working temperature of the training set (μ) T Training set working temperature standard deviation σ T The average operating current of the training set, μ I Training set operating current standard deviation σI :

[0118]

[0119] The training and test sets were then standardized to obtain the standardized training set luminescence intensity matrix I. ELtrain '、Standardized training set working result vector T' train 'Standardized training set working current vector I' train 'Standardized test set luminescence intensity matrix I' ELtest 'Standardized test set working temperature vector T' test 'and standardized test set operating current vector I test ', its expression is as follows:

[0120]

[0121]

[0122] For the standardized training set, the partial least squares regression algorithm is used to construct the luminescence intensity matrix I of the standardized training set of independent variables. ELtrain 'Working vector T of the standardized training set train The relationship between them: First, calculate the first principal component t1 of the independent variable, its expression is:

[0123]

[0124] Where w1 is the coefficient vector of the first principal component of the independent variable, and λ1 is the matrix I. ELtrain ′ T T Ttrain ′、T train ′ T I ELtrain The largest eigenvalue of ′;

[0125] Subsequently, I was established. ELtrain '、T train The regression equation between t1 and t2 is as follows:

[0126]

[0127] Where p1 is the regression coefficient vector, r1 is the regression coefficient, and I ELtrain1 'and T train1 'These are the residual matrices of the equations;

[0128] The expression for solving the regression coefficients is:

[0129]

[0130] Regress the equation residuals again to find the principal component t of the i-th independent variable. iPrincipal component coefficient vector w i , regression coefficient vector p i and regression coefficient r i The expression is:

[0131]

[0132]

[0133]

[0134]

[0135] By selecting the first k principal components to establish a mapping function model, the standardized luminescence intensity vector I of any sample can be obtained. EL 'Regression results with standardized working temperature T' reg 'Working junction temperature function model:'

[0136]

[0137] And to obtain the normalized luminescence intensity vector I of any sample EL 'Regression results with standardized operating current I' reg 'Working current function model between:'

[0138]

[0139] To verify the accuracy of the model, the luminescence intensity matrix I of the standardized test set was used. ELtest Substituting into the working junction temperature function model, we obtain the standardized working junction temperature regression result vector T for the test set. test_reg ', the standardized test set luminescence intensity matrix I ELtest Substituting into the operating current function model, we obtain the standardized operating current regression result vector I for the test set. test_reg Denormalizing both yields the working junction temperature regression vector T. test_reg and the working current regression result vector I test_reg Its expression is:

[0140]

[0141] Calculate the mean absolute error (MAE) of the working junction temperature of the test set. Ttest and the mean absolute error of the operating current (MAE) Itest :

[0142]

[0143] If the mean absolute error does not meet the requirements (threshold requirement), the principal component dimension k of the model is increased, and the regression values ​​are recalculated until a mapping model whose mean absolute error meets the requirements is selected.

[0144] (3) Online monitoring: Based on the trained mapping model and the electroluminescence intensity vector measured by the spectral measurement unit, the operating junction temperature and operating current of the silicon carbide power field-effect transistor are calculated online. Specifically, during the actual operation of the silicon carbide power device, the device's luminescence intensity vector I is obtained online through the spectral measurement unit. EL Convolution smoothing and normalization are performed in the junction temperature and current monitoring unit to obtain the normalized luminous intensity vector I. EL Substituting the values ​​into the partial least squares model, the standardized working junction temperature regression result T is calculated. reg 'Regression results with standardized operating current I' reg Finally, the online operating junction temperature and operating current of the device are obtained through denormalization.

[0145] This embodiment tests one phase of a 1200V / 20A three-phase silicon carbide power device. The main circuit always operates in CCM mode, with a 15mF energy storage capacitor C and a 2.19mH inductor L to ensure that ripple does not affect the measurement. The circuit output voltage is maintained at 25V, the device operating current I ranges from 10 to 26A, the device turn-off voltage is 50 to 300V, and the device junction temperature T ranges from 44.8 to 153.2℃. The device operating current and junction temperature are calibrated using a Hall sensor and a fiber optic thermometer, respectively. The device switching frequency is set to 50kHz, and the dead time is 1 microsecond.

[0146] A total of 1120 data samples were collected, of which 840 were used for model training. The Savitzky-Golay convolution smoothing order was set to 3, and the window size was 11. The principal component counts for the junction temperature regression model and the current regression model were set to 8 and 9, respectively. The working junction temperature regression results for 280 test samples are as follows: Figure 6 As shown. From Figure 6 The results showed that the average absolute error of junction temperature regression was 3.11℃ and the average absolute error of current regression was 0.81A, indicating that the method proposed in this invention has high accuracy.

[0147] In summary, this invention, based on the electroluminescence effect of silicon carbide power devices, obtains high-resolution electroluminescence spectra of the devices through a spectral measurement unit equipped with a spectroscopic dispersion structure and a high-speed CMOS sensor. These spectra are then converted into electrical signals and input to a junction temperature and current monitoring unit for calculating junction temperature and current. The method of this invention utilizes a training dataset obtained from calibration measurements and combines it with a partial least squares regression algorithm to establish mapping models between the electroluminescence spectrum and the device's operating junction temperature and current. These models can be applied online to monitor the operating junction temperature and current of silicon carbide power devices, and have the advantages of high detection bandwidth and small errors in current and junction temperature.

Claims

1. A silicon carbide power field effect transistor junction temperature and current monitoring system, characterized by: The silicon carbide power field effect transistor S1 and S2; The main circuit unit is connected with the silicon carbide power field effect transistor S1 and S2; The driving unit is connected with the silicon carbide power field effect transistor S1 and S2, and is used for providing a switching control signal; The temperature control unit is connected with the package of the silicon carbide power field effect transistor S2, and is used for adjusting the ambient temperature; The junction temperature sampling unit is in contact with the chip of the silicon carbide power field effect transistor S2, and is used for calibrating the actual working junction temperature in the calibration test link; The current sampling unit is connected to the loop of the main circuit unit, and is used for calibrating the actual working current in the calibration test link; The spectrum measurement unit is arranged on the surface of the silicon carbide power field effect transistor S2, and is used for converting the complete electroluminescence spectrum of the silicon carbide power field effect transistor S2 into an electric signal; The junction temperature and current detection unit is connected with the junction temperature sampling unit, the current sampling unit and the spectrum measurement unit, and is used for calculating the junction temperature and the current of the silicon carbide power field effect transistor; The main circuit unit comprises a direct current voltage source Vin, an inductor L, an energy storage capacitor C and a load resistor R; the positive electrode of the direct current voltage source Vin is connected with the drain electrode of the silicon carbide power field effect transistor S1, the negative electrode of the direct current voltage source Vin is connected with the source electrode of the silicon carbide power field effect transistor S2, one end of the energy storage capacitor C and one end of the load resistor R; the source electrode of the silicon carbide power field effect transistor S1 is connected with the drain electrode of the silicon carbide power field effect transistor S2 and one end of the inductor L; the other end of the inductor L is connected with the other end of the energy storage capacitor C and the other end of the load resistor R; The driving unit comprises two silicon carbide power field effect transistor driving circuits and a direct current voltage source for supplying power to the driving circuits; one driving circuit is connected with the gate electrode and the source electrode of the silicon carbide power field effect transistor S1, and the other driving circuit is connected with the gate electrode and the source electrode of the silicon carbide power field effect transistor S2; The spectrum measurement unit comprises an incident slit, a blazed grating and a CMOS line array image sensor; the spectrum measurement unit is fixed on the surface of the silica gel layer of the silicon carbide power field effect transistor S2, and prevents ambient stray light from entering the device package; the incident slit is arranged vertically to the edge of the chip of the silicon carbide power field effect transistor S2, and the electroluminescence light beam of the silicon carbide power field effect transistor S2 passes through the incident slit and is dispersed by the blazed grating, and the light of different wavelengths is converged into different pixels of the CMOS line array image sensor.

2. The silicon carbide power field effect transistor junction temperature and current monitoring system of claim 1, wherein: The junction temperature and current detection unit is internally provided with an FPGA / DSP / single-chip microcomputer, and the FPGA / DSP / single-chip microcomputer is used for calculating the junction temperature and the current.

3. The method of silicon carbide power field effect transistor junction temperature and current monitoring system of claim 1 or 2, wherein: The method comprises the following steps: (1) calibration test link: in the calibration test link, the electroluminescence intensity vector, the actual working junction temperature and the actual working current of the device under various operating conditions are stored, and a data set is obtained; (2) model establishment link: in the model establishment link, the data set of the calibration test link is preprocessed, then the mapping models of the electroluminescence intensity vector and the actual working junction temperature and the actual working current are respectively established based on a partial least squares regression algorithm, and the mapping models are trained by using the data set. (3) Online monitoring link: based on the trained mapping model and the electroluminescence intensity vector measured by the spectral measurement unit, the working junction temperature and the working current of the silicon carbide power field effect transistor are calculated online.

4. The method of a silicon carbide power field effect transistor junction temperature and current monitoring system of claim 3, wherein: The data set is expressed as follows: ； In the formula: represents a light emission intensity matrix; represents the light emission intensity of the first waveband in the spectrum under the first working condition, , ; represents a working junction temperature vector, represents the actual working junction temperature under the first working condition; represents a working current vector; represents the actual working current under the first working condition.

5. The method of a silicon carbide power field effect transistor junction temperature and current monitoring system of claim 4, wherein: The pretreatment is as follows: First, Savitzky-Golay convolution smoothing is applied to the luminescence intensity data to effectively preserve spectral features while removing spectral measurement noise. The convolution smoothing window is set to [value missing]. The order is Then for the first The first of the various working condition samples luminous intensity of band Using the luminescence intensity of adjacent bands , … , polynomial fitting values replace The fitted polynomial is: ； wherein , ,..., are fitting coefficients; Thus obtained The system of linear equations in matrix form is: ； wherein , ,..., is the fitting residual; The above formula is written as: ； The least squares solution for the matrix of fitting coefficients is: ； Further, the polynomial fitting value of :​ ； For each sample, the first to the last convolve the smoothed intensity of the waveband, i.e. obtain a denoised intensity matrix : ； Then randomly sample from the de-noised dataset The group samples constitute a training set, containing a training set luminous intensity matrix , a training set working junction temperature vector , and a training set working current vector , and the expression is: ； Also get the remainder Test set consisting of group samples, containing test set luminous intensity matrix Test set operating junction temperature vector And test set operating current vector : 。 6. The method of a silicon carbide power field effect transistor junction temperature and current monitoring system of claim 5, wherein: The training step of the mapping model using the data set is as follows: Standardize the training set and test set data based on the training set sample distribution: first calculate the training set luminescence intensity mean , training set luminescence intensity standard deviation , training set working junction temperature mean , training set working junction temperature standard deviation , training set working current mean , training set working current standard deviation : ； The training set and the test set are then standardized to obtain a normalized training set luminous intensity matrix , a normalized training set junction temperature vector , a normalized training set operating current vector , a normalized test set luminous intensity matrix , a normalized test set junction temperature vector , and a normalized test set operating current vector , the expressions of which are as follows: ； ； For the standardized training set, a partial least squares regression algorithm is used to establish the relationship between the intensity matrix of the training set and the working junction temperature vector of the training set For the standardized training set, a partial least squares regression algorithm is used to establish the relationship between the intensity matrix of the training set and the working junction temperature vector of the training set For the standardized training set, a partial least squares regression algorithm is used to establish the relationship between the intensity matrix of the training set and the working junction temperature vector of the training set , whose expression is ； wherein is a vector of first principal component coefficients of the independent variables, is a matrix , , , is the largest eigenvalue of the matrix Subsequently established , , Regression equation between the ； wherein, is a vector of regression coefficients, is a regression coefficient, and are equation residual matrices, respectively; The regression coefficient solving expression is: ； The regression is performed again on the equation residual to obtain the principal component of the dependent variable principal component coefficient vector regression coefficient vector and regression coefficient , and the expression is: ； ； ； ； Selecting the front The main component establishes a mapping function model, and obtains the normalized luminous intensity vector of any sample The working junction temperature function model between the regression results of the normalized working junction temperature The working junction temperature function model between the regression results of the normalized working junction temperature ； and deriving a normalized luminescence intensity vector for any sample and a working current function model between the normalized working current and the regression result and a working current function model between the normalized working current and the regression result 。 7. The method of a silicon carbide power field effect transistor junction temperature and current monitoring system of claim 6, wherein: The training further includes verifying accuracy of the trained mapping model by applying the normalized test set light intensity matrix to the junction temperature function model to obtain a test set normalized junction temperature regression result vector , applying the normalized test set light intensity matrix to the junction current function model to obtain a test set normalized junction current regression result vector , and de-normalizing both to obtain a junction temperature regression result vector and a junction current regression result vector , expressed as: ; calculated test set operating junction temperature mean absolute error and operating current mean absolute error : ； If the average absolute error does not meet the requirement, the principal component dimension of the model is increased , the regression value is recalculated until the mapping model with the average absolute error meeting the requirement is selected.

Images