System and method for diagnosis and life estimation of power MOSFET (Metal-Oxide-Semiconductor Field Effect Transistor)

Abstract

The invention discloses a power MOSFET diagnosis and life estimation system and method, and the system comprises a data collection module which is used for configuring a degradation test mode, carrying out the degradation test according to the degradation test mode to obtain degradation data, and carrying out the analysis through the degradation data, and obtaining a potential fault precursor; and the diagnosis module is used for checking and diagnosing the health condition of the internal structure of the power MOSFET according to the fault precursor, and estimating the residual life of the power MOSFET by using an artificial intelligence algorithm. The method is easy to implement, can detect multiple different fault modes at the same time, and regularly, easily and conveniently inspects the health condition of the internal structure of degraded equipment.

Description

Technical Field

[0001] The present invention belongs to the field of electronic technology, and more specifically relates to a system and method for diagnosing and estimating the life of a power MOSFET. Background Art

[0002] Metal oxide silicon field effect transistor, commonly known as MOSFET, is an electronic device used to switch or amplify voltage in circuits. Power MOSFET is a specific type of MOSFET that is specially designed to handle high power applications. The degradation of power MOSFET occupies an important position among the key factors affecting the reliability of power electronic circuits. Usually, the failure location of the degraded power MOSFET is inside the package and cannot be seen from the outside, which makes it difficult to study its failure mode and mechanism. Although there are several power MOSFET failure prediction methods and systems on the market, these methods and systems may not be able to detect and predict failures within the device's internal structure and / or package.

[0003] The Chinese patent document with the document number 107315138B discloses a method and test system for predicting the failure and health of a power MOSFET. The method and system include the following steps: selecting a MOSFET device as a test sample, testing the characteristic parameters of the three parameters of the drain-source voltage, drain current, and device case temperature of a normal device during the aging process, analyzing the test data and establishing a priori degradation model E based on the characteristic parameter Rds(on), determining the failure threshold, and then training the model E using the test data of the actual device under test, correcting the characteristic parameters of the model, predicting the remaining life of the MOSFET device by calculation, outputting the health assessment result, and performing health treatment on the device. However, the method and system are based on a physical model to predict the failure and health of the MOSFET. In order to develop an accurate model, this may require a basic knowledge of the MOSFET wafer degradation mechanism and assembly-related parameters. It also requires a large amount of data on the device's operating conditions, stress history, and material properties. These data may not always be available or difficult to obtain in practice. In the absence of a clearly defined physical model to describe the degradation, predicting the failure of a complex system in practical applications may be challenging.

[0004] Korean patent document No. 102452596B1 discloses a device and method for diagnosing a metal oxide field effect transistor (MOSFET). The device is a device for diagnosing a MOSFET installed on a charging or placing electrical path of a battery pack, for controlling the conduction of a charging or discharging current, and is electrically connected to a gate terminal and a source terminal of the MOSFET. The processor diagnoses whether the MOSFET is faulty by comparing a potential difference measurement value with a normal potential difference value stored in advance. However, the device and method do not disclose a life estimation of a power supply device that may provide an early warning to a user.

[0005] The US patent document with document number 20230194593 discloses a switch device in an alternating fuel transfer pipeline and a method for estimating the remaining service life of the device. The method includes: estimating health characteristics during the use of the switch device, calculating a health state estimation matrix by modeling the degradation of the health characteristics, determining the health state using the health state estimation matrix, and predicting and sending a notification to an output device. However, this method may not be able to widely cover potential fault parameters in different aspects, because all possible failure modes may not be captured by measuring only one or two parameters.

[0006] In the prior art, undetected or unpredicted internal structural failures may increase unexpected machine downtime and maintenance costs due to sudden failures. In addition, different power MOSFET failure modes can be detected by different failure precursors, however, current fault prediction methods and systems may not be able to detect multiple different failure modes at the same time. Therefore, there is a need for an improved method and system that can understand and detect different failure modes during the degradation of power MOSFETs. Summary of the invention

[0007] The purpose of the present invention is to overcome the above-mentioned shortcomings and provide a system and method for diagnosing and estimating the life of a power MOSFET, which can solve the technical problem that the prior art cannot simultaneously detect multiple different failure modes.

[0008] In order to achieve the above object, the technical solution adopted by the present invention is: a system for diagnosing and estimating the life of a power MOSFET, characterized in that it includes:

[0009] A data acquisition module, used to configure a degradation test mode, perform a degradation test according to the degradation test mode to obtain degradation data, and use the degradation data for analysis to obtain potential fault precursors;

[0010] The diagnostic module is used to check and diagnose the health status of the internal structure of the power MOSFET according to the fault precursor, and estimate the remaining life of the power MOSFET using an artificial intelligence algorithm.

[0011] Furthermore, the degradation test mode in the data acquisition module is a power cycle test, and the degradation data is obtained by an accelerated life test method.

[0012] Furthermore, the power cycling test includes indirectly determining the junction temperature by measuring a thermosensitive electrical parameter.

[0013] Furthermore, the data acquisition module includes a comparison submodule for comparing the potential fault precursor with a predetermined fault threshold value to provide an early fault warning.

[0014] Furthermore, the potential precursors to failure include: drain-source on-resistance, gate threshold voltage, diode forward voltage, zero gate-drain current, drain-source breakdown voltage, turn-on voltage, input capacitance, output capacitance and reverse capacitance.

[0015] Furthermore, the diagnostic module uses thermal transient measurement and scanning acoustic microscopy to perform image thermal detection.

[0016] Furthermore, the artificial intelligence algorithm includes a long short-term memory machine learning model.

[0017] Furthermore, the long short-term memory machine learning model is further trained and evaluated.

[0018] Another object of the present invention is to provide a method for diagnosing and estimating the life of a power MOSFET, characterized in that it comprises the following steps:

[0019] Configure a degradation test mode, and perform a degradation test according to the degradation test mode to obtain degradation data;

[0020] Using the degradation data to perform analysis to obtain potential precursors to failures;

[0021] The health status of the internal structure of the power MOSFET is checked and diagnosed based on the fault precursors, and the remaining life of the power MOSFET device is estimated using an artificial intelligence algorithm.

[0022] Furthermore, the step of obtaining the degradation data further comprises the following steps:

[0023] Perform power cycling tests on power MOSFETs and measure thermally sensitive electrical parameters to determine junction temperature for power cycling tests;

[0024] The power cycling test is repeated with high current and high temperature stress, followed by a relaxation period at low temperature and low current.

[0025] Further, the method of estimating the remaining life of a power MOSFET device using an artificial intelligence algorithm comprises the following steps:

[0026] Perform data preparation and preprocessing to extract predictive features for forecasting and lifespan estimation;

[0027] Remove noise and normalize degraded data;

[0028] The normalized degradation data is input into an artificial intelligence algorithm for calculation, wherein the artificial intelligence algorithm includes a long short-term memory machine learning model.

[0029] Furthermore, the noise removal includes: using a moving average filter.

[0030] Furthermore, the long short-term memory machine learning model includes:

[0031] Train using training data;

[0032] Predict the next value of the power MOSFET precursor;

[0033] Evaluate the performance of neural network algorithms.

[0034] Further, the training using training data includes the following steps: using adaptive matrix optimization to adjust the learning rate of each neural network in model training.

[0035] Further, the use of adaptive matrix optimization comprises the following steps:

[0036] Dropout regularization during model training; and

[0037] The corresponding batch size and number of single training rounds are preset according to the size of the degradation data and training data.

[0038] Furthermore, the batch size is 32, and the number of single training rounds is 50.

[0039] Furthermore, it also includes the step of simultaneously updating the output feedback to the long short-term memory machine learning model.

[0040] Compared with the prior art, the present invention has the following advantages:

[0041] The implementation is simple, and includes: a data acquisition module, which is used to configure the degradation test mode, perform degradation test according to the degradation test mode to obtain degradation data, and use the degradation data for analysis to obtain potential fault precursors; a diagnosis module, which is used to check and diagnose the health status of the internal structure of the power MOSFET according to the fault precursors, and use artificial intelligence algorithms to estimate the remaining life of the power MOSFET. Due to the use of artificial intelligence (AI) algorithms that can determine different failure modes during the degradation process of the power MOSFET, the performance and reliability of the final product are improved. Regularly and easily check the health status of the internal structure of the degraded equipment. Potential faults can be warned in advance to allow proactive maintenance and replacement of faulty equipment, thereby reducing unexpected machine downtime and maintenance costs. It is suitable for power MOSFET health diagnosis and life estimation for different types of power equipment, thereby covering a wide range of application areas requiring power switching and control. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 A flow chart showing a method for diagnosing and estimating the life of a power MOSFET in the present invention is shown;

[0043] Figures 2a to 2b The thermal transient measurement curves are shown, showing the progressive curve of die attach layer degradation inside the DUT1 package and the intact curve of DUT10 with no degradation in the die attach layer;

[0044] Figures 3a to 3c Results for die attach layer delamination, lead degradation, and healthy samples are shown, respectively;

[0045] Figures 4a to 4b The fault precursors with predictable characteristics between the body diode voltage (Vsd) and the on-resistance (Rds,on) and the aging power cycle in the lifetime estimation are shown respectively;

[0046] Figures 5a to 5c An example of a failure precursor prediction result of a failure sample is shown;

[0047] Figures 6a to 6c The prediction results of training and validation losses for zero gate-drain current (Idss), on-resistance (Rds,on) and mid-body diode voltage (Vsd) are shown. DETAILED DESCRIPTION

[0048] As required, specific embodiments of the present invention are disclosed herein. However, it should be understood that the disclosed embodiments are only examples of the present invention, which can be implemented in a variety of different forms. Therefore, the specific structural and functional details disclosed herein should not be interpreted as limitations, but only as the basis of the claims. It should be understood that the drawings and their detailed description are not used to limit the present invention to the specific forms disclosed herein. On the contrary, the present invention covers all modifications, equivalents and alternatives that fall within the scope defined by the claims. As used throughout this application, the word "may" means optional (i.e., it means possible), rather than mandatory (i.e., it means must). Similarly, the words "include" and "comprise" are meant to include but are not limited to. In addition, unless otherwise mentioned, the word "one" means "at least one" and the word "multiple" means one or more. When using abbreviations or technical terms, these refer to the generally accepted meanings known in the art.

[0049] The present invention relates to a power metal oxide semiconductor field effect transistor diagnosis and life estimation system, comprising: a data acquisition module with degradation test settings and potential fault precursor measurement; a diagnosis module with thermal transient measurement and scanning acoustic microscope; wherein an artificial intelligence (AI) algorithm is used to estimate the remaining life of a power MOSFET device.

[0050] In a preferred embodiment of the present invention, a power metal oxide semiconductor field effect transistor (MOSFET) diagnosis and life estimation system includes: a data acquisition module with a configuration degradation test and measurement of potential failure precursors; a diagnostic module with thermal transient measurement and scanning acoustic microscopy; wherein an artificial intelligence (AI) algorithm is used to estimate the remaining life of the power MOSFET device.

[0051] In a preferred embodiment of the present invention, the degradation test in the data acquisition module is a power cycle test, which is an accelerated life test (ALT) method to obtain degradation data. The thermal sensitive electrical parameter (TSEP) is measured to indirectly determine the junction temperature to configure and perform the power cycle test.

[0052] In a preferred embodiment of the present invention, the measurement of potential fault precursors in the data acquisition module is configured to measure the potential fault precursors during the power cycle test and compare them with a predetermined fault threshold to provide an early fault warning.

[0053] In a preferred embodiment of the present invention, potential fault precursors include: drain-source on-resistance (Rds,on), gate threshold voltage (Vth), diode forward voltage (Vsd), zero gate drain current (Idss), drain-source breakdown voltage (V(br)dss), turn-on voltage (Von), input capacitance (Ciss), output capacitance (Coss) and reverse capacitance (Crss).

[0054] In a preferred embodiment of the present invention, the diagnostic module utilizes thermal transient measurement and scanning acoustic microscopy (SAM) image thermal detection to inspect and diagnose the health of the internal structure (die attach layer and leads) of the power MOSFET device.

[0055] In a preferred embodiment of the present invention, the artificial intelligence (AI) algorithm includes a long short-term memory (LSTM) machine learning model (110), wherein the LSTM machine learning model is further trained and evaluated.

[0056] The present invention further establishes a power metal oxide semiconductor field effect transistor diagnosis and life estimation method (100), comprising the following steps: obtaining degradation data by performing degradation testing; measuring potential failure precursors to provide early failure warning (106); using thermal transient measurement and scanning acoustic microscope (SAM) images to inspect and diagnose the health of the internal structure of the power MOSFET device (108); repeating power cycle testing and measuring potential failure precursors until the precursors reach a corresponding failure threshold; and using an artificial intelligence (AI) algorithm to estimate the remaining life of the power MOSFET device.

[0057] In a preferred embodiment of the present invention, the acquisition of degradation data further includes the following steps: performing a power cycle test on the power MOSFET device in the degradation test and measuring a thermal sensitive electrical parameter (TSEP) to determine a junction temperature (102) for the power cycle test; and repeating the power cycle test (104) cycle with high current and high temperature stress, followed by a relaxation period at a lower temperature and current.

[0058] In a preferred embodiment of the present invention, the estimation of remaining life includes the following steps: performing data preparation and preprocessing to extract predictable features for prediction and life estimation (109); removing noise and normalizing the degradation data; and inputting the normalized degradation data into an artificial intelligence (AI) algorithm; wherein the AI ​​algorithm includes a long short-term memory (LSTM) machine learning model (110).

[0059] In a preferred embodiment of the present invention, the method further comprises the steps of: training an LSTM machine learning model (112) using training data; predicting the next value of the power MOSFET precursor (114); and evaluating the performance of the neural network algorithm.

[0060] In a preferred embodiment of the present invention, the removal of noise further comprises the following steps: using a moving average filter (MAF) to avoid noise in the degraded data.

[0061] In a preferred embodiment of the present invention, the training of the model includes the following steps: using adaptive moment optimization (Adam) to optimize and adjust the learning rate of each neural network in the model training.

[0062] In a preferred embodiment of the present invention, the method further comprises the following steps: performing dropout regularization during model training; and presetting the corresponding batch size and single training round number according to the size of the degraded data and the training data.

[0063] In a preferred embodiment of the present invention, the batch size is 32 and the number of single training rounds is 50 to prevent overfitting.

[0064] The values ​​of batch size and epoch size are defined based on the prediction results. It can be generalized to a range. For example, the batch size can be determined based on the size of the dataset. The size of the dataset for prediction based on precursor data is not very large, so a batch size between 32 and 128 is acceptable. Keeping the batch size small is a known method to prevent overfitting.

[0065] Similarly, the number of single training epochs was set to 50 to strike a balance between achieving satisfactory performance and not consuming too much time based on the typical dataset size for this type of prediction.

[0066] In a preferred embodiment of the present invention, the method further includes the step of simultaneously updating the output feedback to the LSTM machine learning model.

[0067] As a specific embodiment, the flowchart of the diagnosis and life estimation of the power MOSFET in the present invention is shown in FIG. Figure 1 The method includes the steps of: acquiring degradation data; and measuring a thermal sensitive electrical parameter (TSEP) to determine a junction temperature of a power cycling test device under constant junction temperature conditions (102). Using a drain-source saturation voltage whose voltage is approximately linearly dependent on the junction temperature as the TSEP of the power MOSFET at a small constant measurement current, the method further includes the step of running a power cycling test (104). However, the user may choose to use other methods to acquire degradation data without determining or measuring TSEP.

[0068] The method further includes the following steps: measuring potential failure precursors to provide early failure warning (106); using thermal transient measurements and SAM images to inspect and diagnose the health of the internal structure of the MOSFET device (108); repeating power cycling testing and measuring potential failure precursors until the precursors reach a corresponding failure threshold. The power cycling test includes high current and high temperature stress, followed by a relaxation period at lower temperature and current. The power cycling test provides a realistic simulation of device operation because it subjects the MOSFET to the same type of stress as during normal use, such as temperature and current cycling. The potential failure precursors include Rds,on, Vth, Vsd, Idss, V(br)dss, Von, Ciss, Coss and Crss. These potential failure precursors are acquired and measured at the initial stage and after a given power cycle, wherein these potential failure precursors are selected to represent the key IV and CV characteristics of the power MOSFET for comprehensive diagnosis. The power cycling interval length is designed to ensure that data drift can be accurately captured and avoid omissions. Therefore, the present invention can identify different failure modes according to the corresponding changes of the failure precursors during the degradation process of the power MOSFET, thereby improving the performance and reliability of the product.

[0069] The present invention regularly performs thermal transient measurements and SAM image acquisition to check and diagnose the health of the internal structure of the MOSFET device and correlate it with the changes in the electrical parameters of the degraded device. This is because the fault location of a degraded power MOSFET is usually located inside the package and cannot be observed from the outside, which makes it difficult to study the failure mode and mechanism. The thermal transient measurement shows the changes in the cumulative thermal capacitance and thermal resistance along the heat flow path from the chip to the substrate. Figure 2 shows the thermal transient measurement curve, where Figure 2a The curves show the progression of the die attach layer degradation inside the DUT1 package, while Figure 2b The curve shows a good DUT10 with no die attach degradation. As the power cycling test continues, the curve shifts to the right, indicating an increase in the thermal resistance of the die attach layer. The SAM image shows any delamination or voids that occur within the package. Figure 3a and Figure 3b The results of die attach layer peeling and lead degradation are shown, where Figure 3c A healthy sample is shown. Therefore, the non-destructive inspection method with thermal transient measurement and SAM image helps users understand the physical changes that occur inside the package during the sample degradation process.

[0070] If the precursor does not exceed the failure threshold of 20% of the initial value, the power cycle test will be repeated; if the precursor exceeds the failure threshold, data preparation and preprocessing will be performed. In the present invention, these precursors are considered to be failure precursors because the precursors have predictable characteristics that can be used for life estimation. Figure 4a and Figure 4b Precursor failures with predictable characteristics are shown for body diode voltage (Vsd) and on-resistance (Rds,on) versus aging power cycles, respectively, for lifetime estimation.

[0071] The method further includes preliminary screening of degradation data exceeding the fault threshold, removing noise and normalizing. At this stage, the normalized data can be input into the AI ​​algorithm of the LSTM machine learning model. However, in the present invention, a simple moving average filter (MAF) is further used, where k = 3, to avoid noise in the original data using the following equation:

[0072]

[0073] The present invention uses an LSTM algorithm to process time series data. 30 data points are used to predict the next data value. The LSTM model configuration considers using two hidden layers, with 128 and 64 neurons respectively, and an output layer. The method further includes training the LSTM model and estimating the remaining life of the power MOSFET device. In model training, Adam is used for optimization, and the learning rate is adaptive according to each neural network weight. A common disadvantage of neural network algorithms, especially LSTM, is overfitting. Therefore, a dropout regularization of 0.2, a single training round of 50, and a batch size of 32 are used in the present invention to prevent overfitting. A summary of the LSTM model architecture is listed in Table 1.

[0074] Table 1: Summary of LSTM model architecture

[0075] Model Number of neurons Optimizer Training loss function throw away Activation Function LSTM (128,64) Adam Mean square error 0.2 relu

[0076] For example, the prediction results of the fault precursors of the fault samples are as follows: Figures 5a to 5c As shown, it includes zero gate drain current (Idss) prediction, on-resistance (Rds,on) prediction and mid-body diode voltage (Vsd) prediction. The present invention further evaluates the performance of the proposed algorithm by calculating the mean square error (MSE) of the prediction results. The calculation formula is as follows:

[0077]

[0078] Figures 6a to 6c The performance prediction results of training and validation losses of Idss, Rds,On and Vsd are shown respectively.

[0079] The power MOSFET diagnosis and life estimation system and method of the present invention are very useful for the maintenance of power systems. It regularly monitors the internal health status based on the collected fault precursor data, provides early warnings of potential faults, and captures abnormal behavior before catastrophic failures occur. In addition, by predicting the fault precursor data with the help of the machine learning algorithm used in the present invention, the life of the power MOSFET can be estimated, thereby realizing proactive maintenance and replacement of faulty equipment. The LSTM algorithm can persist degradation information and is very suitable for capturing nonlinear relationships between input and output. This makes data-based models simpler and easier to develop than models based on physical models. Therefore, unexpected machine downtime and maintenance costs can be reduced.

[0080] In addition, the system and method for diagnosis and life estimation of power MOSFET in the present invention are not limited to a specific type of power MOSFET. The data-driven machine learning method in the present invention makes it possible to apply it to different types of power devices, including but not limited to MOSFET, IGBT, SiC MOSFET, etc. The active power cycle test in the present invention can reproduce the working conditions for different power devices through the similarity in its working principle, such as thermal stress under high power. Since the electrical characteristics and internal structure of the power devices are similar, these methods are universal. The machine learning algorithm is able to process time series data of different electrical parameters and devices, so the user can input any data that is considered to have predictable characteristics into the model for prediction. Therefore, the system and method of the present invention can cover a wide range of application fields that require power switches and controls, including but not limited to automobiles, aerospace, power generation and telecommunications.

[0081] The above explanation of the present invention is not limited to the aforementioned embodiments and drawings, and it is obvious to those skilled in the art that various substitutions, modifications and changes may be made without departing from the scope of the present invention.

Claims

1. A system for diagnosing and estimating the life of a power MOSFET, characterized in that: include: A data acquisition module, used to configure a degradation test mode, perform a degradation test according to the degradation test mode to obtain degradation data, and use the degradation data for analysis to obtain potential fault precursors; The diagnostic module is used to check and diagnose the health status of the internal structure of the power MOSFET according to the fault precursor, and estimate the remaining life of the power MOSFET using an artificial intelligence algorithm.

2. The system for diagnosing and estimating the life of a power MOSFET according to claim 1, characterized in that: The degradation test mode in the data acquisition module is a power cycle test, which acquires degradation data by an accelerated life test method.

3. The system for diagnosing and estimating the life of a power MOSFET according to claim 2, characterized in that: The power cycling test involves indirectly determining the junction temperature by measuring a thermosensitive electrical parameter.

4. The system for diagnosing and estimating the life of a power MOSFET according to claim 1, characterized in that: The data acquisition module includes a comparison submodule, which is used to compare the potential fault precursor with a predetermined fault threshold to perform an early fault warning.

5. The system for diagnosing and estimating the life of a power MOSFET according to claim 4, characterized in that: The potential fault precursors include: drain-source on-resistance, gate threshold voltage, diode forward voltage, zero gate drain current, drain-source breakdown voltage, turn-on voltage, input capacitance, output capacitance and reverse capacitance.

6. The system for diagnosing and estimating the life of a power MOSFET according to claim 1, characterized in that: The diagnostic module uses thermal transient measurement and scanning acoustic microscopy to perform image thermal detection.

7. The system for diagnosing and estimating the life of a power MOSFET according to claim 1, characterized in that: The artificial intelligence algorithm includes a long short-term memory machine learning model.

8. The system for diagnosing and estimating the life of a power MOSFET according to claim 1, characterized in that: The long short-term memory machine learning model was further trained and evaluated.

9. A method for diagnosing and estimating the life of a power MOSFET, characterized in that: The following steps are involved: Configure a degradation test mode, and perform a degradation test according to the degradation test mode to obtain degradation data; Using the degradation data to perform analysis to obtain potential precursors to failures; The health status of the internal structure of the power MOSFET is checked and diagnosed based on the fault precursors, and the remaining life of the power MOSFET device is estimated using an artificial intelligence algorithm.

10. The method for diagnosing and estimating the life of a power MOSFET according to claim 9, characterized in that: The obtaining of degradation data further comprises the following steps: Perform power cycling tests on power MOSFETs and measure thermally sensitive electrical parameters to determine junction temperature for power cycling tests; The power cycling test is repeated with high current and high temperature stress, followed by a relaxation period at low temperature and low current.

11. The method for diagnosing and estimating the life of a power MOSFET according to claim 9, characterized in that: The method of estimating the remaining life of a power MOSFET device using an artificial intelligence algorithm comprises the following steps: Perform data preparation and preprocessing to extract predictive features for forecasting and lifespan estimation; Remove noise and normalize degraded data; The normalized degradation data is input into an artificial intelligence algorithm for calculation, wherein the artificial intelligence algorithm includes a long short-term memory machine learning model.

12. The method for diagnosing and estimating the life of a power MOSFET according to claim 11, characterized in that: The noise removal includes using a moving average filter.

13. The method for diagnosing and estimating the life of a power MOSFET according to claim 11, characterized in that: The long short-term memory machine learning model includes: Train using training data; Predict the next value of the power MOSFET precursor; Evaluate the performance of neural network algorithms.

14. The method for diagnosing and estimating the life of a power MOSFET according to claim 13, characterized in that: The training using training data includes the following steps: using adaptive matrix optimization to adjust the learning rate of each neural network in model training.

15. The method for diagnosing and estimating the life of a power MOSFET according to claim 11, characterized in that: The adaptive matrix optimization comprises the following steps: Dropout regularization during model training; and The corresponding batch size and number of single training rounds are preset according to the size of the degradation data and training data.

16. The method for diagnosing and estimating the life of a power MOSFET according to claim 15, characterized in that: The batch size is 32, and the number of single training rounds is 50.

17. The method for diagnosing and estimating the life of a power MOSFET according to claim 11, characterized in that: It also includes the step of simultaneously updating the output feedback to the long short-term memory machine learning model.

Images