Method and system for junction temperature measurement and lifetime estimation of power devices

Abstract

The application provides a junction temperature measurement and life estimation method and system of a power device. The method comprises: obtaining actual operation data of the power device, obtaining an actual junction temperature fluctuation curve of the power device in actual operation with time variation according to the actual operation data and based on a thermal network model of the power device; using a rain flow method to implement online actual cycle number statistics on the actual junction temperature fluctuation curve and extract actual cycle junction temperature fluctuation amplitudes and actual cycle junction temperature means under each actual cycle number; estimating the life of the power device based on the actual cycle junction temperature fluctuation amplitudes and the actual cycle junction temperature means under each actual cycle number; inputting the actual cycle junction temperature fluctuation amplitudes and the actual cycle junction temperature means under each actual cycle number into a pre-trained neural network model every predetermined time, and outputting current thermal resistance parameter values of the thermal network model; and calibrating the thermal network model based on the current thermal resistance parameter values for junction temperature estimation in the next stage.

Description

Technical Field

[0001] This application relates to the field of power device technology, and in particular to a method and system for measuring junction temperature and predicting lifetime of power devices. Background Technology

[0002] With the accelerated electrification of transportation, especially the rapid development of electric vehicles, third-generation wide-bandgap SiC (silicon carbide) power inverters, featuring higher voltage ratings, faster switching speeds, and higher power densities, are gradually replacing traditional Si (silicon)-based power inverters. In the automotive field, power inverters, as key components of electric drive systems, present particularly significant reliability challenges. Compared to traditional Si-based IGBT (Insulated Gate Bipolar Transistor) power devices, SiC-based MOSFETs (Metal-Oxide-Semiconductor Field-Effect Transistors) offer superior performance, but their higher power density leads to more demanding heat dissipation requirements and a more severe reliability challenge.

[0003] Statistics show that over 90% of electric drive system failures originate from fatigue of power devices in inverters. Domestic and international research indicates that the root cause of SiC MOSFET fatigue failure lies in the inconsistent thermal expansion coefficients at the interfaces of different packaging materials, leading to thermal stress failure under junction temperature fluctuations. Excessively high junction temperatures are the primary cause of over 50% of power device failures, and the probability of failure doubles for every 10°C increase in ambient temperature. Increased junction temperature not only causes deviations in thermistor parameters such as threshold voltage, on-state voltage drop, and switching time, but also accelerates the degradation of the thermal and electrical conductivity of electrode materials. In practical applications, power device failures are often triggered by either excessively high junction temperatures or periodic fluctuations: the former may lead to instantaneous device failure, while the latter accelerates the aging process of the packaging.

[0004] It is worth noting that aging phenomena such as chip solder layer fatigue during use can significantly alter thermal resistance, causing a gradual increase in the deviation between the initially calibrated thermal network model parameters and the actual values ​​during use. If the thermal resistance model is not corrected in time, the junction temperature estimation error will accumulate over time, potentially leading to a failure to provide timely warnings when power devices experience thermal failure. Therefore, real-time and accurate junction temperature measurement and lifetime prediction of power devices in automotive inverters are crucial for preventing thermal failure and optimizing SiC power inverter design. Summary of the Invention

[0005] The purpose of this application is to provide a method and system for measuring the junction temperature and estimating the lifetime of power devices, which can solve at least one of the technical problems mentioned in the prior art.

[0006] One aspect of this application provides a method for measuring the junction temperature and predicting the lifetime of a power device. The method includes: acquiring actual operating data of the power device during actual operation; obtaining an actual junction temperature fluctuation curve of the power device over time based on the actual operating data and a pre-established thermal network model of the power device; performing online actual cycle counts on the actual junction temperature fluctuation curve using the rainflow method and extracting the actual cycle junction temperature fluctuation amplitude and the average actual cycle junction temperature at each actual cycle count; predicting the lifetime of the power device based on the actual cycle junction temperature fluctuation amplitude and the average actual cycle junction temperature at each actual cycle count; inputting the actual cycle junction temperature fluctuation amplitude and the average actual cycle junction temperature at each predetermined time interval into a pre-trained neural network model and outputting the current thermal resistance parameter value of the thermal network model; calibrating the thermal network model based on the current thermal resistance parameter value for use in the next stage of junction temperature prediction.

[0007] Further, the life prediction of the power device based on the actual cyclic junction temperature fluctuation amplitude and the average actual cyclic junction temperature under each actual cycle number includes: establishing a fatigue life model of the power device, wherein the fatigue life model is a function with the cyclic junction temperature fluctuation amplitude and the average cyclic junction temperature as independent variables and the maximum number of fatigue cycles that can be withstood under the cyclic junction temperature fluctuation amplitude and the average cyclic junction temperature as dependent variables; inputting the actual cyclic junction temperature fluctuation amplitude and the average actual cyclic junction temperature under each actual cycle number into the fatigue life model, and outputting the maximum number of fatigue cycles that can be withstood under the actual cyclic junction temperature fluctuation amplitude and the average actual cyclic junction temperature corresponding to each actual cycle number; predicting the remaining life of the power device based on each actual cycle number and the maximum number of fatigue cycles that can be withstood under the actual cyclic junction temperature fluctuation amplitude and the average actual cyclic junction temperature corresponding to each actual cycle number.

[0008] Further, predicting the remaining life of the power device based on the actual number of cycles and the maximum number of fatigue cycles that it can withstand under the actual cycle junction temperature fluctuation amplitude and the average actual cycle junction temperature corresponding to each actual cycle number includes: predicting the remaining life of the power device based on the actual number of cycles and the maximum number of fatigue cycles that it can withstand under the actual cycle junction temperature fluctuation amplitude and the average actual cycle junction temperature corresponding to each actual cycle number; obtaining the fatigue damage accumulated by the power device after the previous fatigue cycles according to Miner's rule; and obtaining the remaining life of the power device based on the fatigue damage accumulated by the power device after the previous fatigue cycles.

[0009] Furthermore, establishing the fatigue life model of the power device includes: establishing the fatigue life model of the power device based on the Coffin-Manson life model, wherein the maximum number of fatigue cycles that the device can withstand under the cyclic junction temperature fluctuation amplitude and the average cyclic junction temperature are constants. Circulating junction temperature fluctuation amplitude The power is directly proportional to the power, and the maximum number of fatigue cycles and activation energy are also proportional. It is related to an exponential function of the ratio of the mean cyclic junction temperature. It is a constant. , and The value was obtained through pre-experiment fitting.

[0010] Furthermore, the aforementioned , and The value is obtained through pre-experiment fitting, including: using a power device as the test object, performing switching operations on the power device to make the power device alternately heated and cooled; using the cyclic junction temperature fluctuation amplitude and the average cyclic junction temperature as the control variables of the experiment; acquiring multiple sets of first test data under the same cyclic junction temperature fluctuation amplitude and different average cyclic junction temperatures; acquiring multiple sets of second test data under different cyclic junction temperature fluctuation amplitudes and the same average cyclic junction temperature, wherein one set of test data in the multiple sets of first test data and the multiple sets of second test data is the same; and fitting the multiple sets of first test data and the multiple sets of second test data to obtain the value. , and The value of .

[0011] Furthermore, the data obtained by fitting multiple sets of the first experimental data and multiple sets of the second experimental data... , and The values ​​include: obtained by fitting multiple sets of the first experimental data. The value was obtained by fitting multiple sets of the second experimental data. The value; in the known and After obtaining the value, the results were obtained by fitting multiple sets of the first experimental data and multiple sets of the second experimental data. The value of .

[0012] Furthermore, the method further includes: initially calibrating the thermal resistance parameter values ​​in the thermal network model; conducting a pre-test on the entire life cycle of the power device, acquiring historical operating data of the power device during the pre-test process, obtaining a historical junction temperature fluctuation curve of the power device over time based on the historical operating data, and calibrating the thermal resistance parameter in the thermal network model based on the historical operating data during the pre-test process to obtain the thermal resistance parameter calibration value of the thermal network model during the pre-test process; using the rainflow method to perform online historical cycle count statistics on the historical junction temperature fluctuation curve and extracting the historical cycle junction temperature fluctuation amplitude and the historical cycle junction temperature mean value under each historical cycle count; using the historical cycle junction temperature fluctuation amplitude and the historical cycle junction temperature mean value under each historical cycle count as input, and the thermal resistance parameter of the thermal network model as output, using a loss function to calculate the error between the output predicted value of the thermal resistance parameter and the thermal resistance parameter calibration value to train the neural network model, so as to obtain the finally trained neural network model.

[0013] Furthermore, the power device includes a power chip, a heat sink, and a multilayer structure located between the power chip and the heat sink. The method further includes: establishing a thermal network model of the power device based on the Foster model, wherein the order of the thermal network model is determined according to the number of layers of the multilayer structure, and the thermal network model is related to the reference temperature at the heat sink, power consumption, junction temperature, and thermal resistance of each order of the power device during use.

[0014] Furthermore, the initial calibration of the thermal resistance parameters in the thermal network model includes: using the power consumption calculated during use, the measured reference temperature at the heat sink base plate, and the junction temperature of the power device as known quantities, and using the thermal resistance of each order in the thermal network model as unknown quantities, to initially calibrate the thermal resistance parameter values ​​in the thermal network model.

[0015] Another aspect of this application provides a system for measuring the junction temperature and estimating the lifetime of a power device. The system includes a memory, a processor, and a computer program stored in the memory. The processor executes the computer program to implement the steps of the power device junction temperature measurement and lifetime estimation method as described above.

[0016] The junction temperature measurement and lifetime prediction method and system for power devices according to one or more embodiments of this application first performs fatigue cycle statistics on the junction temperature fluctuation curve obtained from the data of the entire life cycle using the rainflow method, extracts the cyclic junction temperature fluctuation amplitude and the cyclic junction temperature mean at each cycle number, and uses only the cyclic junction temperature fluctuation amplitude and the cyclic junction temperature mean at each cycle number as the input of the neural network model, thereby greatly improving the accuracy of model training under the premise of limited model computing power.

[0017] The junction temperature measurement and lifetime prediction method and system of power devices in one or more embodiments of this application construct an aging feedback system through a neural network to achieve adaptive calibration of the thermal network model of the power device, and correct the problem that the thermal resistance parameter value of the thermal network model gradually deviates from the initial calibration value due to the aging of the power device.

[0018] The junction temperature measurement and lifetime prediction method and system for power devices according to one or more embodiments of this application obtains historical operating data of the power device during the entire lifetime of the power device through pre-testing. Based on the historical operating data, a historical junction temperature fluctuation curve of the power device over time during the pre-testing process is obtained. During the pre-testing process, the thermal resistance parameters in the thermal network model are calibrated using the historical operating data to obtain the calibrated values ​​of the thermal resistance parameters of the thermal network model during the pre-testing process. Then, the rainflow method is used to perform online historical cycle counts on the historical junction temperature fluctuation curve and extract the historical cycle junction temperature fluctuation amplitude and the historical cycle junction temperature average value at each historical cycle number. Then, using the historical cycle junction temperature fluctuation amplitude and the historical cycle junction temperature average value at each historical cycle number as inputs and the thermal resistance parameters of the thermal network model as outputs, the error between the predicted value of the output thermal resistance parameters and the calibrated value of the thermal resistance parameters is calculated using a loss function to train the neural network model. The junction temperature measurement and lifetime prediction method for power devices in this application simulates the entire lifetime cycle of the power device through pre-testing, and can obtain historical test data of the power device throughout the aging process (i.e., under all operating conditions), covering the entire lifetime cycle data of the power device from the early stage to the end of its lifetime. Therefore, the original operating data used to train the neural network model of the junction temperature measurement and lifetime prediction method of power devices in this application is more comprehensive, and thus, the accuracy of the neural network model of this application is higher.

[0019] The junction temperature measurement and lifetime prediction method and system of power devices in one or more embodiments of this application can complete the thermal resistance calibration and junction temperature estimation without the need for a complex finite element simulation model that requires high theoretical knowledge. In addition, during use, there is no need to equip additional circuits to measure temperature-sensitive electrical parameters such as threshold voltage. Existing negative temperature coefficient thermistor sensors can be used without additional circuits, and negative temperature coefficient thermistor sensors are cheaper.

[0020] The junction temperature measurement and lifetime prediction method and system for power devices in one or more embodiments of this application employ a neural network method to avoid the lack of an accurate theoretical model between cycle lifetime and thermal resistance, and to ensure that the mapping relationship between the two is accurate. Attached Figure Description

[0021] Figure 1 This is a flowchart of the preliminary calibration process for a power device junction temperature measurement and lifetime prediction method according to an embodiment of this application.

[0022] Figure 2 This is a layered schematic diagram of a SiC power device.

[0023] Figure 3 This is a schematic diagram of a thermal network model of a power device based on the Foster model, according to an embodiment of this application.

[0024] Figure 4 This is a flowchart of the sub-steps of statistical fatigue cycle using the rainflow method according to one embodiment of this application.

[0025] Figure 5 This is a schematic diagram illustrating the formation of a closed load loop according to one embodiment of this application.

[0026] Figure 6 This is a schematic diagram of a neural network model according to an embodiment of this application.

[0027] Figure 7 This is a flowchart illustrating the junction temperature measurement and lifetime prediction method for a power device according to an embodiment of this application during the actual operation of the power device.

[0028] Figure 8 This is a schematic diagram illustrating the principle of a fatigue life model parameter fitting test according to an embodiment of this application.

[0029] Figure 9 This is a schematic diagram of the junction temperature fluctuation of a power device during an experiment according to one embodiment of this application.

[0030] Figure 10 Fitting an embodiment of this application A schematic diagram.

[0031] Figure 11 Fitting an embodiment of this application A schematic diagram.

[0032] Figure 12 Fitting an embodiment of this application A schematic diagram.

[0033] Figure 13 This is a schematic block diagram of a power device junction temperature measurement and lifetime prediction system according to an embodiment of this application. Detailed Implementation

[0034] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses consistent with some aspects of this application as detailed in the appended claims.

[0035] There are generally two methods for estimating the junction temperature of power devices: one is to establish a thermal resistance network using a finite element simulation model; the other is to measure temperature-sensitive parameters such as threshold voltage, turn-on delay time, and turn-off delay time using the temperature-sensitive electrical parameter method to obtain the junction temperature. However, the method of establishing a thermal resistance network using a finite element simulation model requires advanced theoretical knowledge and is relatively complex; while the method of measuring using temperature-sensitive electrical parameters requires the introduction of additional circuitry, and some temperature-sensitive electrical parameters are also affected by aging, thus causing measurement errors as well.

[0036] In addition, there is a lack of accurate theoretical models relating the cycle life and thermal resistance of power devices, and traditional theoretical models cannot accurately predict the degree of change in thermal resistance at a certain stage of the life.

[0037] In view of this, this application provides an improved method and system for measuring junction temperature and estimating lifetime of power devices, which can effectively solve at least one technical problem existing in the above-mentioned related technologies.

[0038] This application constructs an aging feedback system using a neural network to achieve adaptive calibration of the thermal network model of power devices, further enabling online junction temperature estimation and lifetime prediction. Compared to traditional methods that rely on offline models, this application effectively solves the problem of thermal network model accuracy decay during the aging process of power devices, and achieves dynamic compensation of thermal resistance parameters, real-time junction temperature estimation, and remaining lifetime prediction.

[0039] The method and system for measuring junction temperature and predicting lifetime of power devices according to this application will be described in detail below with reference to the accompanying drawings. Unless otherwise specified, the features of the following embodiments and implementations can be combined with each other.

[0040] Figure 1 This document discloses a flowchart of the preliminary calibration process for a power device junction temperature measurement and lifetime prediction method according to an embodiment of this application. Figure 1 As shown, the junction temperature measurement and lifetime prediction method for power devices according to one embodiment of this application may include steps S101 to S105 in the early calibration process.

[0041] In step S101, a thermal network model of the power device is established.

[0042] Power devices include a power chip, a heat sink, and a multilayer structure located between the power chip and the heat sink. The power chip may, for example, be a SiC power chip. Figure 2 A layered schematic diagram of a SiC power device 200 is shown. (For example...) Figure 2 As shown, the SiC power device 200 is composed of multiple layers of materials stacked together, including a chip 201, an upper solder layer 202, an upper copper layer 203, a ceramic substrate 204, a lower copper layer 205, a lower solder layer 206, an iron substrate 207, a substrate connection layer 208, and a heat sink 209. The thermal expansion coefficients of the materials in different layers are different. When the device is turned on, the chip generates heat, causing the junction temperature to rise, which is then transferred layer by layer to the heat sink, resulting in a significant temperature gradient distribution inside the power device. When the power device reaches thermal equilibrium, the temperature gradient of each layer of material will remain relatively stable.

[0043] In some embodiments, a thermal network model of the power device can be established based on the Foster model. The order of the thermal network model can be determined based on the number of layers in the multilayer structure. The thermal network model is correlated with the reference temperature at the heat sink of the power device during operation. Power consumption Junction temperature of power devices and thermal resistance of each order related.

[0044] Figure 3 A schematic diagram of a thermal network model of a power device based on the Foster model, according to an embodiment of this application, is shown. Figure 3 As shown, the thermal network model corresponds to formula (1), which can be simplified to formula (2) in the steady state, as shown below: (1) (2) in, This refers to the junction temperature of the power device; This is the reference temperature at the heat dissipation base plate that can be measured by a negative temperature coefficient thermistor sensor during use. Power loss (or simply power consumption) of power devices during use. The thermal resistance is the i-th order. ; Let i be the heat capacity of the i-th order; Let be the order of the heat network model. Figure 3 The heat network model is shown as having an order of 3, i.e. .

[0045] If expressed in matrix form, then: (3) in: The state matrix; For control matrix; This is the output matrix; It is a feedforward matrix; For input; This is the output, i.e., the junction temperature; For state variables ( (Represents a vector with 1 row and 3 columns).

[0046] Thermal resistance of each order With heat capacity The following relationships also exist: (4) in, The thermal conductivity of the material; Specific heat capacity of the material; The density of the material; The thickness is the material thickness.

[0047] Power consumption of power devices Including switching power consumption With conduction power consumption ,Right now: (5) Switching power consumption Including the energy lost during the process. With the energy lost due to shutdown All of these can be found in the chip datasheet, and the expressions are as follows: (6) in, This refers to the switching frequency of the power device.

[0048] On-power consumption For a short period of time Internal conduction voltage drop and current The mean of the integral of the product is shown below: (7) Continue to refer to Figure 1 In step S102, the thermal resistance parameter values ​​in the thermal network model established in step S101 are initially calibrated.

[0049] After establishing a thermal network model in the form of formulas (1)-(7) above, the thermal resistance parameter values ​​in the thermal network model are initially calibrated. During the calibration process, the power consumption calculated during use is used as the reference. The measured reference temperature at the heat sink base. and junction temperature of power devices As known quantities, the thermal resistances of each order in the thermal network model As an unknown quantity.

[0050] Among them, the power consumption of power devices during use The reference temperature at the heat sink base plate can be calculated based on the three-phase voltage and current measured by voltage and current sensors, and the reference temperature can be measured by a negative temperature coefficient temperature sensor. Infrared temperature sensors were used to measure the junction temperature inside the power device after the outer layer was removed. By collecting the transient impedance curve of a single pulse, the thermal resistance of each order can be obtained by fitting it according to formulas (1)-(7), thus completing the thermal resistance calculation. Initial offline calibration of parameter values.

[0051] In step S103, a pre-test is conducted on the power device throughout its entire lifespan to obtain historical operating data during the pre-test process. This historical operating data includes values ​​for multiple parameters in the thermal network model, including but not limited to voltage, current, and the junction temperature of the power device. Etc. Junction temperature The acquisition method is the same as the offline calibration above. For example, an infrared temperature sensor can be used to measure the junction temperature inside the power device after removing the outer layer. Considering that the thermal resistance of power devices changes with aging, it is necessary to calibrate the thermal resistance parameters in the thermal network model using historical operating data during the pre-testing process. Offline calibration is frequently used during pre-testing; specifically, it requires using the reference temperature at the heat sink base. Power consumption and junction temperature of power devices As a known quantity, the thermal resistance of each order in the thermal network model The calibration is performed as an unknown quantity, including the reference temperature at the heat sink base. The power consumption can be measured using a negative temperature coefficient temperature sensor. The three-phase voltage and current, measured by voltage and current sensors, can be calculated. This is used to obtain the junction temperatures of the power devices during the pre-test process. Subsequently, the junction temperature of the power device during the pre-test process can be obtained. Historical junction temperature fluctuation curve over time.

[0052] In step S104, the rainflow method is used to perform online historical cycle count statistics on the historical junction temperature fluctuation curve and extract the historical cycle junction temperature fluctuation amplitude and the historical cycle junction temperature average value at each historical cycle count.

[0053] The following will combine Figure 4 This section details how to use the rainflow method for fatigue cycle statistics.

[0054] Figure 4 A flowchart illustrating the sub-steps of statistical fatigue cycle using the rainflow method in one embodiment of this application is disclosed. Figure 4 As shown, the specific steps are as follows: Step S401: Initialize the data buffer.

[0055] Create a buffer with 3 locations to store junction temperatures: location 1 stores historical junction temperature values. Location 2 stores the latest junction temperature value. Position 3 receives the next new junction temperature estimate. .

[0056] Step S402: Turning point detection.

[0057] examine Does it constitute and The turning point between them, i.e., the judgment Is it a local peak / valley?

[0058] Turning point condition: If yes arrive Extreme points in the path, if < > or > < If ., then the condition is met.

[0059] Step S403: Data update and cyclic feature recording.

[0060] Non-turning point processing: If It does not constitute a turning point, using value overwrite and retain Continue reading the next data point as .

[0061] Handling turning points: If It's a turning point, using value overwrite ,use value overwrite ; and record separately Further examination of the most recent individual records Check if any of the four values ​​form a closed load cycle. If a cycle is formed, record the cycle characteristics (mark the cycle number and the corresponding cycle junction temperature fluctuation amplitude). Average cyclic junction temperature And remove related data points from individual records.

[0062] The following will combine Figure 5 This will explain how to calculate and form a cycle. Figure 5 A schematic diagram illustrating the formation of a closed load loop is shown, representing one embodiment of this application. Figure 5 As shown, due to If it's a turning point, it will be recorded separately, and over time, it will form a composite of numerous... The sequence consists of peaks and valleys. This sequence presents a pattern of "peak-valley-peak-valley" or "valley-peak-valley-peak".

[0063] Taking the "peak-valley-peak-valley" form as an example, if the second value is greater than or equal to the fourth value, or the third value is less than or equal to the first value, then a closed loop exists.

[0064] For example, Figure 5 The four shown on the left This forms a closed loop, where the average junction temperature of the loop is... = (second value + third value) / 2; Cyclic junction temperature fluctuation amplitude = (third value - second value) / 2.

[0065] In subsequent columns, delete the second and third values, keeping only the first and fourth values, such as... Figure 5 As shown on the right, if new ones are generated subsequently... So, put the new Place it at the end of the queue.

[0066] For the “valley-peak-valley-peak” pattern: a closed loop is formed if the second value is less than or equal to the fourth value, or the third value is greater than or equal to the first value.

[0067] Step S404: Step iteration loop.

[0068] Repeat steps S402 and S403 above to continuously process the new input data stream, dynamically update the buffer, and extract the load loop. The cumulative record is... That is, the amplitude of the cyclic junction temperature fluctuation is recorded. and average junction temperature The number of times the loop has already occurred .

[0069] return Figure 1 In step S105, a neural network is used to fit the relationship between fatigue cycles and thermal resistance parameters of the thermal network model during the aging process. The historical cycle junction temperature fluctuation amplitude and the average historical cycle junction temperature (i.e., the values ​​obtained from the rainflow method) are used to calculate the relationship. The thermal resistance parameter of the thermal network model is used as the input of the neural network model, and the thermal resistance parameter of the thermal network model is used as the output of the neural network model. The error between the predicted value of the thermal resistance parameter output by the neural network model and the calibration value of the thermal resistance parameter obtained in step S103 is calculated by the loss function Loss to train the neural network model, so as to obtain the final trained neural network model. Figure 6 A schematic diagram illustrating a neural network model according to an embodiment of this application is shown. Figure 6 As shown, the two-dimensional data is first constructed from the historical cycle junction temperature fluctuation amplitude and the historical cycle junction temperature mean at each historical cycle number. (For example, 100 rows and 3 columns, derived from accumulated experimental data; the more iterations, the more accumulated experimental data.) This data is input into the neural network model, representing the two-dimensional input data. Flattened into one dimension, the output is obtained by passing it through three fully connected layers. .in, , , These are the results obtained after passing through three fully connected layers; , , and This is the weight matrix. , , . and The bias is obtained through training. , and The activation function is ReLU; The activation function is softmax.

[0070] Figure 7 This document discloses a flowchart illustrating the junction temperature measurement and lifetime prediction method for a power device according to an embodiment of this application during the actual operation of the power device. For example... Figure 7 As shown, the junction temperature measurement and lifetime prediction method for a power device according to one embodiment of this application may include steps S701 to S705 in the actual operation of the power device.

[0071] In step S701, the actual operating data of the power device during actual operation is obtained, and based on the actual operating data and the thermal network model, the actual junction temperature fluctuation curve of the power device during actual operation is obtained.

[0072] The operation process of step S701 is similar to that of step S103 above.

[0073] In step S702, the rainflow method is used to perform online statistics on the actual number of cycles of the actual junction temperature fluctuation curve and extract the actual cyclic junction temperature fluctuation amplitude and the average actual cyclic junction temperature at each actual number of cycles.

[0074] The operation process of step S702 is similar to that of step S104 above.

[0075] In step S703, the lifespan of the power device is estimated based on the actual cyclic junction temperature fluctuation amplitude and the average actual cyclic junction temperature for each actual number of cycles.

[0076] In some embodiments, step S703 may further include steps S801 to S803.

[0077] Cyclic junction temperature fluctuation amplitude and average cyclic junction temperature The fatigue life is the most important factor in predicting fatigue life. Therefore, in step S801, a fatigue life model of the power device is established. The fatigue life model is a function with the amplitude of the cyclic junction temperature fluctuation and the average cyclic junction temperature as independent variables, and the maximum number of fatigue cycles that can be withstood under the amplitude of the cyclic junction temperature fluctuation and the average cyclic junction temperature as the dependent variable.

[0078] In some embodiments, the fatigue life model of a power device can be established based on the Coffin-Manson lifetime model. This model includes the maximum number of fatigue cycles that the device can withstand under varying junction temperature fluctuations and average junction temperatures, along with a constant. Circulating junction temperature fluctuation amplitude The power is directly proportional to the power, and the maximum number of fatigue cycles and activation energy are also proportional. It is related to an exponential function of the ratio of the mean cyclic junction temperature. It is a constant.

[0079] In one embodiment, the fatigue life model of a power device can be expressed as follows: (8) in, Boltzmann's constant; This represents the amplitude of the cyclic junction temperature fluctuation. This is the average junction temperature over the cycle; This indicates that the amplitude of the cyclic junction temperature fluctuation is... The average junction temperature of the circulation is The maximum number of fatigue cycles that the device can withstand. , and The value can be obtained through pre-experiment fitting.

[0080] Figure 8 A schematic diagram illustrating the principle of a fatigue life model parameter fitting test according to an embodiment of this application is shown. Figure 8 As shown, for example, using a single SiC MOSFET half-bridge as the test object, the cyclic junction temperature fluctuation amplitude is used as the test object. and average junction temperature As the control variable in the experiment, Figure 9 A schematic diagram illustrating the junction temperature fluctuation of a power device during an experiment according to an embodiment of this application is shown. (Referring to...) Figure 8 and Figure 9 As shown, the power device is switched on and off to alternately heat and cool it. A square wave signal controls the switch to cycle through the high-power current source for 2 seconds to heat the power device and then shut off for 4 seconds to cool it. The gate voltage is always on, and the junction temperature is measured in real time using an infrared temperature sensor. When the power device is turned on by the switch, it is in the heating process, and the junction temperature gradually increases over time. When the power device is turned off by the switch, it is in the cooling process, and the junction temperature gradually decreases over time. This cycle is repeated tens of thousands of times until the power device's lifespan is exhausted. The average junction temperature and the fluctuation amplitude of the cycle are differentiated among the groups by adjusting the drive current and cooling water flow rate. To ensure the specified junction temperature waveform, the current source current and the on / off time should be adjusted periodically during the cyclic test.

[0081] During the experiment, the junction temperature fluctuation amplitude under the same cycle was obtained. Average junction temperature of different cycles Multiple sets of first-stage test data were obtained; the junction temperature fluctuation amplitude under different cycles was acquired. Average junction temperature of the same cycle The data consists of multiple sets of second experimental data, where one set of data from each of the multiple sets of first and second experimental data is identical. Then, a fitting algorithm can be used to obtain... . 、 and The value of .

[0082] Figure 10 The fitting of one embodiment of this application is disclosed A schematic diagram. (See example.) Figure 10 As shown, it can be based on multiple sets of first test data (i.e., the amplitude of cyclic junction temperature fluctuations). Average junction temperature under constant and different cycles The following experimental data were fitted to obtain The value of . Figure 11 The fitting of one embodiment of this application is disclosed A schematic diagram. (See example.) Figure 11 As shown, it can be based on multiple sets of second experimental data (i.e., at the average cyclic junction temperature). Constant junction temperature fluctuation amplitude under different cycles The following experimental data were fitted to obtain The value of . Figure 12 The fitting of one embodiment of this application is disclosed A schematic diagram. (See example.) Figure 12 As shown, in the known and After obtaining the value, the data were fitted based on all the data from multiple sets of first experimental data and multiple sets of second experimental data. The value of .

[0083] In step S802, the actual cyclic junction temperature fluctuation amplitude and the average actual cyclic junction temperature under each actual cycle number are input into the fatigue life model, and the maximum number of fatigue cycles that can be withstood under the actual cyclic junction temperature fluctuation amplitude and the average actual cyclic junction temperature corresponding to each actual cycle number are output.

[0084] In step S803, the remaining lifespan of the power device is predicted based on the actual number of cycles, the actual cycle junction temperature fluctuation amplitude corresponding to each actual cycle number, and the maximum number of fatigue cycles that the device can withstand under the average actual cycle junction temperature.

[0085] In some embodiments, step S803 may further include steps S811 and S812.

[0086] In step S811, the remaining life of the power device can be predicted based on the actual number of cycles, the actual junction temperature fluctuation amplitude corresponding to each actual number of cycles, and the maximum number of fatigue cycles that the device can withstand under the average actual junction temperature. The fatigue damage accumulated by the power device after the previous fatigue cycles is obtained according to Miner's rule.

[0087] According to Miner's rule, the fatigue damage accumulated by a power device after the previous fatigue cycles can be expressed as: (9) in, For the front The damage accumulated over a fatigue cycle can be presented as a percentage. The average junction temperature corresponding to the i-th fatigue cycle and the amplitude of cyclic junction temperature fluctuation The maximum number of fatigue cycles that can be withstood under certain conditions is calculated using formula (9); The average junction temperature corresponding to the i-th fatigue cycle and the amplitude of cyclic junction temperature fluctuation The number of iterations under the given conditions is calculated by step S702.

[0088] In step S812, the remaining life of the power device can be obtained based on the fatigue damage accumulated by the power device after the previous fatigue cycle.

[0089] Remaining lifespan It can be represented as: (10) When the remaining lifespan When the value drops to 0, it is considered invalid.

[0090] Return to reference Figure 7 In step S704, every predetermined time interval Input the actual cyclic junction temperature fluctuation amplitude and the average actual cyclic junction temperature obtained in step S702 for each actual number of cycles into the trained neural network model, and output the current thermal resistance parameter value of the thermal network model.

[0091] In step S705, the thermal network model is calibrated based on the current thermal resistance parameter value obtained in step S704 for junction temperature prediction in the next stage.

[0092] This application can correlate fatigue cycles with the thermal resistance parameter values ​​of a thermal network model. During use, the thermal resistance parameter values ​​of the thermal network model are calibrated periodically, i.e., at fixed predetermined intervals. Next, the statistics will be obtained using the rainflow method. The input is fed into a trained neural network model, which then outputs corrected thermal resistance parameter values ​​for the thermal network model, to be used over a predetermined period of time. Junction temperature inside This estimation allows for dynamic compensation of the decreased accuracy of thermal resistance parameters in the thermal network model caused by the aging of power devices. (Scheduled time) This refers to the step size for updating the thermal resistance parameter values ​​of the thermal network model.

[0093] Furthermore, the junction temperature measurement and lifetime prediction method for power devices in this application involves conducting a pre-test on the entire lifetime of the power device to obtain historical operating data during the pre-test process. Based on the historical operating data and a thermal network model, a historical junction temperature fluctuation curve of the power device over time during the pre-test process is obtained. Then, the rainflow method is used to perform online historical cycle counts on the historical junction temperature fluctuation curve and extract the historical cycle junction temperature fluctuation amplitude and the historical cycle junction temperature average at each historical cycle number. Finally, the historical cycle junction temperature fluctuation amplitude and the historical cycle junction temperature average at each historical cycle number are used as inputs, and the thermal resistance parameters of the thermal network model are used as outputs to train the neural network model. Therefore, compared with existing methods that use data from the actual operation of power devices to train neural network models, the junction temperature measurement and lifetime prediction method for power devices in this application simulates the entire lifetime cycle of power devices through pre-testing. This allows the acquisition of historical test data of power devices throughout the aging process (i.e., under all operating conditions), covering the entire lifetime cycle data of power devices from the early stage to the end of their lifespan. Therefore, the original operating data used to train the neural network model by the junction temperature measurement and lifetime prediction method for power devices in this application is more comprehensive, and consequently, the accuracy of the neural network model in this application is higher.

[0094] Existing methods typically use raw operating data (such as voltage and current) of power devices directly as input to neural network models. However, due to computational limitations, the neural network model cannot input all voltage and current data from the entire lifespan of the power device; it can only extract a small segment of data. Therefore, the amount of data used for neural network model training is limited, resulting in low training accuracy. In contrast to existing methods that directly use raw operating data, the junction temperature measurement and lifespan prediction method for power devices proposed in this application first uses the rainflow method to perform fatigue cycle statistics on the junction temperature fluctuation curves obtained from the entire lifespan operating data. This extracts the cyclic junction temperature fluctuation amplitude and average cyclic junction temperature at each cycle number. Only the cyclic junction temperature fluctuation amplitude and average cyclic junction temperature at each cycle number are used as input to the neural network model, thereby significantly improving the model training accuracy despite limited computational power.

[0095] The junction temperature measurement and lifetime prediction method for power devices in this application constructs an aging feedback system through a neural network to achieve adaptive calibration of the thermal network model of the power device, thereby correcting the problem that the thermal resistance parameter value of the thermal network model gradually deviates from the initial calibration value due to the aging of the power device.

[0096] The junction temperature measurement and lifetime prediction method for power devices in this application can complete thermal resistance calibration and junction temperature estimation without the need for complex finite element simulation models that require high theoretical knowledge. Furthermore, it does not require additional circuitry to measure temperature-sensitive electrical parameters such as threshold voltage during use. Existing negative temperature coefficient thermistor sensors can be used without additional circuitry, and the price of negative temperature coefficient thermistor sensors is also lower.

[0097] The junction temperature measurement and lifetime prediction method for power devices in this application uses a neural network method to avoid the problem of lack of accurate theoretical model between cycle lifetime and thermal resistance, and ensures that the mapping relationship between the two is accurate.

[0098] This application also provides a system for measuring the junction temperature and predicting the lifetime of power devices. Figure 13 A schematic block diagram of a power device junction temperature measurement and lifetime prediction system 900 according to an embodiment of this application is shown. Figure 13 As shown, a power device junction temperature measurement and lifetime prediction system 900 according to one embodiment of this application includes a processor 901, an internal bus 902, a network interface 903, a memory 904, and a non-volatile memory 905, and may also include other hardware required for business operations. The processor 901 can read the corresponding computer program from the non-volatile memory 905 into the memory 904 and then run it to implement the steps of the power device junction temperature measurement and lifetime prediction method as described above. Of course, in addition to the software implementation, this application does not exclude other implementation methods, such as logic devices or a combination of hardware and software, etc. That is to say, the execution subject of the following processing flow is not limited to each logic unit, but can also be hardware or logic devices.

[0099] The junction temperature measurement and lifetime prediction system 900 for power devices of this application can have similar beneficial technical effects as the junction temperature measurement and lifetime prediction method for power devices described above, therefore, it will not be described in detail here.

[0100] The junction temperature measurement and lifetime prediction method and system for power devices provided in this application have been described in detail above. Specific examples have been used to illustrate the junction temperature measurement and lifetime prediction method and system for power devices in this application. The descriptions of the above embodiments are only for helping to understand the core ideas of this application and are not intended to limit this application. It should be noted that those skilled in the art can make several improvements and modifications to this application without departing from the spirit and principles of this application, and all such improvements and modifications should fall within the protection scope of the appended claims.

Claims

1. A method for junction temperature measurement and lifetime prediction of a power device, characterized in that, include: Obtain actual operating data of the power device during actual operation, and based on the actual operating data and a pre-established thermal network model of the power device, obtain the actual junction temperature fluctuation curve of the power device during actual operation as a function of time. The rainflow method was used to perform online statistics on the actual number of cycles of the actual junction temperature fluctuation curve and to extract the actual cyclic junction temperature fluctuation amplitude and the average actual cyclic junction temperature at each actual number of cycles. The lifespan of the power device is estimated based on the actual cycle junction temperature fluctuation amplitude and the average actual cycle junction temperature under each actual cycle number. At predetermined intervals, the actual cycle junction temperature fluctuation amplitude and the actual cycle junction temperature average value under each actual cycle number are input into a pre-trained neural network model, and the current thermal resistance parameter value of the thermal network model is output. The thermal network model is calibrated based on the current thermal resistance parameter values ​​for junction temperature prediction in the next stage.

2. The method of claim 1, wherein, The method of estimating the lifespan of the power device based on the actual cyclic junction temperature fluctuation amplitude and the average actual cyclic junction temperature for each actual cycle number includes: A fatigue life model for the power device is established, wherein the fatigue life model is a function with the amplitude of the cyclic junction temperature fluctuation and the mean cyclic junction temperature as independent variables, and the maximum number of fatigue cycles that the device can withstand under the amplitude of the cyclic junction temperature fluctuation and the mean cyclic junction temperature as the dependent variable. The actual cyclic junction temperature fluctuation amplitude and the average actual cyclic junction temperature under each actual cycle number are input into the fatigue life model, and the maximum number of fatigue cycles that can be withstood under the actual cyclic junction temperature fluctuation amplitude and the average actual cyclic junction temperature corresponding to each actual cycle number are output. The remaining lifespan of the power device is predicted based on the actual number of cycles, the amplitude of the actual junction temperature fluctuation corresponding to each actual number of cycles, and the maximum number of fatigue cycles that the device can withstand under the average actual junction temperature.

3. The method of claim 2, wherein, The prediction of the remaining lifespan of the power device based on the actual number of cycles, the actual junction temperature fluctuation amplitude corresponding to each actual number of cycles, and the maximum number of fatigue cycles that it can withstand under the average actual junction temperature includes: The remaining life of the power device is predicted based on the actual number of cycles, the actual junction temperature fluctuation amplitude corresponding to each actual number of cycles, and the maximum number of fatigue cycles that the device can withstand under the average actual junction temperature. The fatigue damage accumulated by the power device after fatigue junction temperature cycles is obtained according to Miner's rule. The remaining life of the power device is obtained based on the fatigue damage accumulated by the power device after fatigue junction temperature cycling.

4. The method of claim 2, wherein, The establishment of the fatigue life model for the power device includes: The fatigue life model of the power device is established based on the Coffin-Manson lifetime model, wherein... The maximum number of fatigue cycles and constant that can be withstood under the fluctuation amplitude and average value of the cyclic junction temperature. Circulating junction temperature fluctuation amplitude The power is directly proportional to the power, and the maximum number of fatigue cycles and activation energy are also proportional. It is related to an exponential function of the ratio of the mean cyclic junction temperature. It is a constant. , and The value was obtained through pre-experiment fitting.

5. The method as described in claim 4, characterized in that, The , and The values ​​obtained through pre-experiment fitting include: Using a power device as the test object, the power device is subjected to switching operations to cause the power device to be alternately heated and cooled; The amplitude of the cyclic junction temperature fluctuation and the average cyclic junction temperature were used as the control variables in the experiment. Obtain multiple sets of first test data under the same cycle junction temperature fluctuation amplitude and different cycle junction temperature averages; Multiple sets of second experimental data are obtained under different cycle junction temperature fluctuation amplitudes and the same cycle junction temperature average value, wherein one set of experimental data in the multiple sets of first experimental data and multiple sets of second experimental data is the same; Based on the fitting of multiple sets of the first experimental data and multiple sets of the second experimental data, the following was obtained: , and The value of .

6. The method as described in claim 5, characterized in that, The data obtained by fitting multiple sets of the first experimental data and multiple sets of the second experimental data is as follows. , and The values ​​include: Based on fitting multiple sets of the first experimental data, the following was obtained: The value; Based on fitting multiple sets of the second experimental data, the following was obtained: The value; In the known and After obtaining the value, the results were obtained by fitting multiple sets of the first experimental data and multiple sets of the second experimental data. The value of .

7. The method as described in claim 1, characterized in that, The method further includes: The thermal resistance parameter values ​​in the aforementioned thermal network model are initially calibrated. A pre-test is conducted on the entire life cycle of the power device to obtain historical operating data of the power device during the pre-test process. Based on the historical operating data, a historical junction temperature fluctuation curve of the power device over time is obtained during the pre-test process. The thermal resistance parameters in the thermal network model are calibrated using the historical operating data based on the thermal network model during the pre-test process to obtain the calibration value of the thermal resistance parameters of the thermal network model during the pre-test process. The rainflow method was used to perform online historical cycle count statistics on the historical junction temperature fluctuation curve and extract the historical cycle junction temperature fluctuation amplitude and the historical cycle junction temperature mean value under each historical cycle count. The neural network model is trained by taking the historical cycle junction temperature fluctuation amplitude and the historical cycle junction temperature average at each historical cycle number as inputs, and the thermal resistance parameter of the thermal network model as output. The error between the predicted value of the output thermal resistance parameter and the calibrated value of the thermal resistance parameter is calculated using a loss function to obtain the finally trained neural network model.

8. The method as described in claim 7, characterized in that, The power device includes a power chip, a heat sink, and a multi-layer structure located between the power chip and the heat sink. The method further includes: A thermal network model of the power device is established based on the Foster model. The order of the thermal network model is determined according to the number of layers in the multilayer structure. The thermal network model is related to the reference temperature at the heat sink base, power consumption, junction temperature of the power device, and thermal resistance of each order during use.

9. The method as described in claim 8, characterized in that, The initial calibration of the thermal resistance parameters in the thermal network model includes: The thermal resistance parameter values ​​in the thermal network model are initially calibrated using the power consumption calculated during use, the measured reference temperature at the heat sink base, and the junction temperature of the power device as known quantities, and the thermal resistance of each order in the thermal network model as unknown quantities.

10. A system for measuring the junction temperature and predicting the lifetime of a power device, comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the steps of the method for measuring junction temperature and estimating lifetime of power devices as described in any one of claims 1 to 9.

Images