Welding type IGBT aging on-line prediction method fusing analytic physical model and deep learning

Abstract

The invention discloses a welding type IGBT aging online prediction method fusing an analytic physical model and deep learning, and the method comprises the following steps: collecting temperature-sensitive electrical parameters of an IGBT module in real time, the temperature-sensitive electrical parameters including saturation voltage drop, collector current, shell temperature, current stress and voltage stress; inputting the temperature-sensitive electrical parameters into a pre-trained deep neural network model, and outputting the junction temperature of the IGBT; processing the junction temperature fluctuation data by adopting a dynamically updated online rain flow counting method, and counting the junction temperature fluctuation quantity in an equivalent period in real time; inputting the junction temperature fluctuation quantity into a service life prediction model based on a Coffin-Manson equation, and calculating a failure cycle number; accumulating the damage amount based on a linear accumulated damage theory; when the damage amount reaches a first threshold value, early warning is triggered, when the damage amount reaches a second threshold value, failure is judged, and the remaining service life is predicted according to the current damage amount.

Description

Technical Field

[0001] The present invention relates to the technical field of power semiconductor devices, and in particular to an online prediction method for aging of welded IGBTs that integrates analytical physical models with deep learning. Background Art

[0002] As a core power device in modern power electronics systems, insulated-gate bipolar transistors (IGBTs) are widely used in renewable energy generation, smart grids, electric vehicles, industrial frequency conversion, and other fields. As power electronics devices evolve toward higher power density and higher reliability, the long-term reliability of IGBT modules has become increasingly prominent. In actual operation, IGBT modules are subjected to a long-term coupling of electrical, thermal, and mechanical stresses. Failure modes such as solder layer aging and bond wire detachment have become key factors limiting device lifespan.

[0003] Traditional IGBT aging assessment methods mainly rely on regular offline testing: first, offline testing requires shutdown and disassembly, seriously affecting the normal operation of the equipment; second, there are differences between laboratory accelerated aging conditions and actual operating conditions, resulting in insufficient prediction accuracy; third, most existing models do not consider the individual differences of IGBTs and the dynamic stress changes in actual operation, making it difficult to accurately reflect the true aging status of individual devices.

[0004] With the development of artificial intelligence technology, deep neural networks have shown advantages in the field of condition monitoring, but their application in IGBT life prediction still faces challenges: on the one hand, purely data-driven methods lack physical mechanism support and have insufficient generalization capabilities in small sample cases; on the other hand, existing methods mostly focus on single stress factors and fail to effectively integrate the aging mechanism of electrical-thermal multi-physical field coupling.

[0005] Therefore, developing an aging prediction method that can be implemented online, takes into account the advantages of physical mechanisms and data-driven, and is applicable to different types of IGBTs is of great significance to improving the reliability of power electronic systems and realizing predictive maintenance. Summary of the Invention

[0006] To address the shortcomings of existing technologies, such as low shutdown detection efficiency, poor adaptability to individual differences, and insufficient modeling of multi-physics field coupling mechanisms, the present invention provides an online prediction method for IGBT aging that integrates analytical physical models with deep learning. Its innovative design includes:

[0007] Multi-physics field dynamic fusion mechanism: Real-time monitoring of junction temperature through deep neural network (DNN), with input parameters including saturation voltage drop, collector current, case temperature, current stress and voltage stress, realizing the online quantitative analysis of the aging effect of electrical-thermal-mechanical multi-physics field coupling for the first time.

[0008] Dynamic and collaborative life prediction architecture: Innovatively adopts a dynamically updated online rain flow counting method: It processes junction temperature extremes in real time through dual buffers and dynamically outputs temperature difference fluctuation ΔT, supporting accurate input of physical models;

[0009] Combined with the Coffin-Manson life model: the model parameters (proportional coefficient a, shape parameter b) are updated online every 5000 cycles to solve the problem that traditional static models ignore individual differences and operating condition drift.

[0010] Threshold Life Warning System: Based on Miner's linear cumulative damage theory, a dual-threshold failure judgment mechanism is proposed for the first time:

[0011] When the cumulative damage amount ΣD ≥ 0.8, the warning is triggered (first threshold);

[0012] When ΣD≥1.0, it is judged as failure (second threshold);

[0013] The remaining life is output in real time using the formula RUL=(1-ΣD)×100%.

[0014] The present invention specifically adopts the following technical solutions:

[0015] An online prediction method for welding IGBT aging that integrates analytical physical models and deep learning includes the following steps:

[0016] Real-time acquisition of temperature-sensitive electrical parameters of the IGBT module, including saturation voltage drop, collector current, case temperature, current stress, and voltage stress;

[0017] Inputting the temperature-sensitive electrical parameters into a pre-trained deep neural network model to output the junction temperature of the IGBT;

[0018] Adopting the dynamically updated online rain flow counting method to process the junction temperature fluctuation data, and to count the junction temperature fluctuation amount within the equivalent cycle in real time;

[0019] Inputting the junction temperature fluctuation into a life prediction model based on the Coffin-Manson equation to calculate the number of failure cycles;

[0020] Accumulate damage based on linear cumulative damage theory;

[0021] When the damage amount reaches a first threshold, an early warning is triggered; when it reaches a second threshold, failure is determined, and the remaining service life is predicted based on the current damage amount.

[0022] Furthermore, the online rain flow counting method dynamically updates the temperature extreme value through the dual buffer area and outputs the junction temperature fluctuation magnitude in real time.

[0023] Furthermore, the parameters of the life prediction model are updated every 5000 cycles.

[0024] Furthermore, the deep neural network model includes 3 hidden layers.

[0025] Furthermore, the current stress and voltage stress are collected in real time by a Hall sensor.

[0026] Furthermore, the remaining service life is calculated by the following formula: RUL=(1-ΣD)×100%; wherein ΣD is the accumulated damage amount.

[0027] Furthermore, the first threshold is 0.8, and the second threshold is 1.0.

[0028] And, an online prediction system for welding IGBT aging that integrates analytical physics models and deep learning, including:

[0029] Data acquisition module: real-time acquisition of IGBT temperature-sensitive electrical parameters including saturation voltage drop, collector current, case temperature, current stress, and voltage stress;

[0030] Junction temperature monitoring module: a pre-trained deep neural network model that inputs the temperature-sensitive electrical parameters and outputs the junction temperature;

[0031] Dynamic processing module: uses online rain flow counting method to process junction temperature fluctuation data and outputs junction temperature fluctuation quantity in real time;

[0032] Life prediction module: calculates the number of failure cycles based on the Coffin-Manson equation;

[0033] Damage accumulation module: accumulates damage and triggers warning / failure;

[0034] Output module: predicts the remaining service life; the remaining service life is calculated using the following formula: RUL=(1-ΣD)×100%; where ΣD is the accumulated damage amount.

[0035] And, a computer device includes a processor and a memory, wherein the memory stores a computer program, characterized in that when the program is executed by the processor, the steps of the above method are implemented.

[0036] A non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program implements the steps of the above method when executed by a processor.

[0037] Compared with the prior art, the present invention and its preferred embodiments have at least the following beneficial effects:

[0038] Realize online monitoring of all operating conditions: By integrating physical models with deep learning technology, the IGBT aging status can be obtained in real time without downtime, completely solving the problem of equipment interruption caused by traditional methods relying on offline detection.

[0039] Improve prediction accuracy and adaptability:

[0040] The dynamically updated online rain flow counting method combined with a dual buffer mechanism accurately captures junction temperature fluctuations and significantly optimizes the calculation accuracy of the temperature difference ΔT.

[0041] The dynamic calibration of the Coffin-Manson lifetime model parameters every 5000 cycles effectively overcomes the interference of individual device differences and operating condition drift on the prediction results.

[0042] Innovation lifespan early warning mechanism:

[0043] Damage accumulation judgment based on dual thresholds (0.8 warning / 1.0 failure) enables hierarchical control of the aging process for the first time;

[0044] The remaining life formula intuitively quantifies the module health status and supports proactive maintenance decisions.

[0045] Multi-physics coupling analysis capabilities:

[0046] Deep neural networks simultaneously analyze the interactive effects of current stress, voltage stress, and temperature field, breaking through the limitations of single-factor modeling and more realistically reflecting the electro-thermal-mechanical coupling aging mechanism under actual working conditions.

[0047] Expanding technology ubiquity:

[0048] The dynamic parameter update mechanism is adaptable to different types of IGBT modules, and the method can be seamlessly migrated to high-reliability scenarios such as new energy power generation and electric vehicles. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments:

[0050] Figure 1 This is a flow chart of the online rainflow counting method according to an embodiment of the present invention;

[0051] Figure 2 This is a basic idea diagram of online IGBT life prediction according to an embodiment of the present invention;

[0052] Figure 3 Schematic diagram of a neural network algorithm model for detecting IGBT junction temperature according to an embodiment of the present invention;

[0053] Figure 4 This is a graph showing cycle life prediction results for an embodiment of the present invention;

[0054] Figure 5 This is an overall flow chart of an embodiment of the present invention. DETAILED DESCRIPTION

[0055] Hereinafter, specific embodiments of the present application will be described in detail with reference to the accompanying drawings. Based on these detailed descriptions, those skilled in the art will be able to clearly understand the present application and implement the present application. Without violating the principles of the present application, the features of different embodiments may be combined to obtain new implementations, or certain features of certain embodiments may be substituted to obtain other preferred implementations.

[0056] To make the features and advantages of the present invention more clearly understood, the following embodiments are specifically described in detail with reference to the accompanying drawings.

[0057] This invention discloses an online prediction method for welded IGBT aging that integrates analytical physical models with deep learning. First, the IGBT's saturation voltage drop, collector current, and case temperature data are collected during power cycling experiments to establish a Coffin-Manson-based lifespan model and a DNN neural network junction temperature monitoring model. During actual operation, these electrothermal parameters are collected in real time and input into the DNN model to obtain junction temperature data. An online rain flow counting method is used to process the junction temperature fluctuation curve, calculating the temperature difference ΔT within the equivalent cycle. This data is then input into the lifespan model to calculate the number of cycles to failure, and damage is accumulated based on Miner's linear fatigue cumulative damage theory. When the accumulated damage reaches a threshold, an early warning or failure determination is triggered. This innovative fusion of physical models and deep learning technology not only considers temperature factors but also other aging effects such as current stress and voltage stress. This method can more accurately reflect the aging mechanism of the electrothermal multi-physics coupling under actual operating conditions, addressing the problems of traditional methods such as the need for downtime testing and insufficient prediction accuracy, and significantly improving lifespan prediction accuracy. This method enables online and accurate lifespan prediction for IGBT modules of different models and processes, and can be widely applied to power electronics system health management in fields such as renewable energy power generation and electric vehicles. This method can overcome the shortcomings of different models and individual differences of IGBT modules, realize online IGBT life prediction, and solve the problem of predicting the remaining service life of a single IGBT module in actual application.

[0058] like Figure 1 、 Figure 5 As shown, it is specifically implemented according to the following steps:

[0059] Step 1: Collect temperature-sensitive electrical parameters during the power cycling experiment, pre-process them to create a data set, and establish a DNN neural network junction temperature monitoring model;

[0060] Step 2: Calibrate the Coffin-Manson life model through power cycling experiments.

[0061] Step 3: Input the collected electrothermal parameters into the pre-trained DNN model to obtain the IGBT junction temperature T j ;

[0062] Step 4: Draw a junction temperature fluctuation curve based on the junction temperature data;

[0063] Step 5: Use the online rain flow counting method to process the junction temperature fluctuation data and statistically output the temperature difference ΔT within the equivalent period;

[0064] Step 6: Input ΔT into the life prediction model to calculate the number of failure cycles N f ;

[0065] Step 7: Damage accumulation based on Miner linear fatigue cumulative damage theory;

[0066] Step 8: Predict the remaining useful life of the IGBT module.

[0067] As a preferred solution of this embodiment, in step 1: the DNN neural network includes 3 hidden layers, and the input node is V CE , I C and T C , the output node is the junction temperature T j .

[0068] Preprocessing includes removing abnormal data and normalizing data.

[0069] As a preferred solution of this embodiment, in step 2, the Coffin-Manson device lifetime model is:

[0070]

[0071] N f is the number of cycles to failure; a is the proportionality coefficient; b is the shape parameter, which reflects the effect of device performance on the degree of relationship; is the junction temperature fluctuation.

[0072] Accelerated aging data is obtained through power cycling experiments, and based on the Coffin-Manson equation, two core parameters a (proportional coefficient) and b (shape parameter) are calibrated to establish a life prediction model.

[0073] After the Coffin-Manson lifespan model is established, the parameters are continuously updated through online monitoring data (updated every 5000 cycles) to improve the accuracy of individualized predictions.

[0074] As a preferred solution of this embodiment, in step 5: the online rain flow counting method calculates the characteristic parameters by the following formula:

[0075]

[0076] Where T max and T min are the extreme values ​​of the temperature cycle.

[0077] As a preferred solution of this embodiment, in step 7: the basic theoretical expression of Miner linear fatigue cumulative damage theory is:

[0078] Σ

[0079] ΣD is the cumulative damage (D=1 when failure occurs), n i is the actual number of cycles under the i-th stress condition, N f is the number of cycles to failure under the ith stress condition (calculated by the Coffin-Manson model).

[0080] As a preferred solution of this embodiment, in step 8: predict the remaining service life according to the current damage rate:

[0081] RUL = (1-ΣD)×100%

[0082] When ΣD≥0.8, an early warning is triggered, and when ΣD≥1.0, it is considered to be invalid.

[0083] This paper proposes an online prediction method for IGBT aging that integrates analytical physical models with deep learning. Its main purpose is to provide a high-precision online remaining life prediction solution for IGBT modules. This method analyzes the correlation between junction temperature and IGBT fatigue aging, uses power cycling experimental data to establish an initial life model based on the Coffin-Manson equation, and then uses a deep neural network to accurately monitor junction temperature. An improved online rain flow counting algorithm is used to extract features from junction temperature fluctuation data. Finally, a complete life prediction model is constructed in conjunction with Miner's linear fatigue cumulative damage theory to assess the IGBT aging status in real time. This method innovatively combines physical models with data-driven methods, significantly improving the accuracy and practicality of IGBT life prediction.

[0084] Based on the solution provided in the above embodiment, in the power cycling experiment, a given temperature cycle is repeatedly applied to the module until it fails. The life data of the power cycling test is used to establish a life model based on the Coffin-Manson equation. In the power cycling experiment, the temperature-sensitive electrical parameters of the IGBT are extracted and a DNN neural network junction temperature monitoring model is established. Under actual working conditions, the temperature-sensitive electrical parameter acquisition module is connected to the IGBT to measure the saturation voltage drop, collector current and case temperature of the IGBT. The measured temperature-sensitive electrical parameters are input into the pre-trained DNN neural network junction temperature monitoring model to obtain the junction temperature of the IGBT module. According to the junction temperature obtained by the junction temperature monitoring model, a junction temperature fluctuation curve is drawn. The junction temperature fluctuation data is used as input, and the online rain flow counting method is used to statistically output the temperature difference ∆T and the average value T within the equivalent period. mThe resulting temperature difference, ∆T, is used as input in the online life prediction model to calculate the number of cycles to failure for the power device. Miner's linear fatigue accumulation theorem is used to accumulate the damage identified for each equivalent cycle. This predicts the remaining useful life of the IGBT module.

[0085] The following provides a more specific application example to further demonstrate and introduce the specific implementation and details of the solution of the present invention:

[0086] like Figure 1 The flowchart of the online prediction method for IGBT aging by integrating analytical physical model and deep learning is shown in the present invention, which includes the following steps:

[0087] Step 1: Build a power cycle experimental platform to collect the saturation voltage drop V of the IGBT module CE , collector current I C and shell temperature T C The data is collected and preprocessed. The preprocessing includes: outlier removal and data normalization.

[0088] Step 2: Use the preprocessed IGBT temperature-sensitive electrical parameters to input the trained DNN model to obtain the junction temperature of the IGBT.

[0089] The neural network algorithm implemented in the present invention comprises three hidden layers, three input nodes, and one output node; the three input nodes are respectively saturation voltage drop V CE , collector current I C and shell temperature T C , the output value of the output node is the junction temperature of the IGBT; the neural network algorithm model is shown in the figure Figure 2 As shown;

[0090] Step 3: Draw the junction temperature sequence of the IGBT module based on the current junction temperature obtained in step 2 and the historical junction temperature data.

[0091] Step 4: Use the improved online rain flow counting algorithm to process the junction temperature sequence and statistically output the temperature difference ∆T within the equivalent period.

[0092] After the real-time junction temperature data arrives, the online rain flow counting method first processes the data into extreme value form. The maximum and minimum values ​​are processed using two buffer areas to identify full-wave stress cycles and half-wave stress cycles, and the life prediction model is updated at the same time.

[0093] Figure 3 The figure shows the flow chart of the online rainflow counting method. In the figure, T represents the temperature, and ∆T represents the swing value of the equivalent cycle, that is, the temperature difference. , the calculation formula is:

[0094]

[0095] When the online rainflow counting method identifies a new maximum, if one already exists, the two are compared; otherwise, the new value is directly stored in the stack. If the new value is greater than the first existing maximum, the number of existing minima is checked. If there is only one minimum, a half-wave stress cycle is identified, and the temperature difference ∆T is the difference between the old maximum and the old minimum, replacing the old one. If there are two or more minima, a full-wave stress cycle is identified, and the temperature difference ∆T is the difference between the new minimum and the old maximum, replacing the old one and removing the new minimum. If the new maximum is not greater than the old one, it is stored as the first value in the stack. After this operation is completed, if the existing maximum is not a value, the above operation is repeated.

[0096] When a new temperature extreme value is reached, the system dynamically updates the maximum value (Max-Buffer) and the minimum value (Min-Buffer) through the dual buffer area. If the new maximum value is greater than the current stack first value, then:

[0097] When Min-Buffer has only one value, it is identified as a half-wave cycle, and ΔT = Old Max - Old Min is calculated, and the old Max is replaced by the new Max;

[0098] When Min-Buffer has ≥ 2 values, it is identified as a full-wave cycle, ΔT = New Min - Old Max is calculated, the new Min is removed, and Max is updated.

[0099] If the new maximum value is less than or equal to the first value in the stack, it is stored in the Max-Buffer.

[0100] Step 5: Input the temperature difference ∆T into the lifespan consumption model (Coffin-Manson model) to calculate the number of cycles to failure. The parameters are continuously updated (every 5,000 cycles) using online monitoring data.

[0101] The Coffin-Manson model is widely used in the life prediction of power electronic devices. It is a classic thermomechanical fatigue model specifically used to evaluate the aging failure process of the solder layer of IGBT modules. The Coffin-Manson device life model is:

[0102]

[0103] Where N f is the number of cycles to failure; a is the proportionality coefficient; b is the shape parameter, which reflects the effect of device performance on the degree of relationship; Its outstanding advantages are clear physical meaning, efficient calculation, simple experimental calibration, low cost, and it is particularly suitable for online life prediction under actual working conditions.

[0104] After every 5000 cycles, the system refits the Coffin-Manson parameters a and b using the least squares method based on the ΔT sequence output by the rainflow method.

[0105] Step 6: Calculate the damage amount of each cycle based on Miner's linear cumulative damage theory. The damage of each equivalent cycle identified needs to be accumulated and the total damage amount ΣD is accumulated.

[0106] In a certain period of actual operation of the IGBT module, the number of times a certain junction temperature fluctuation cycle occurs is n1. In the life model, the number of cycles corresponding to the same junction temperature fluctuation cycle is N1. Therefore, the damage caused by this junction temperature fluctuation to the component during this period of operation is D=n1 / N1. Based on this, the module is subjected to a thermal stress level S i Next action n i The damage under the cycle is D i =n i / N i , if there are k thermal stress levels S i Under the action of n i cycles, the total damage can be defined as

[0107]

[0108] ΣD is the cumulative damage (D=1 when failure occurs), n i is the actual number of cycles under the i-th stress condition, N f is the number of cycles to failure under the i-th stress condition.

[0109] Step 7: Predict the remaining useful life of the IGBT module. A warning is triggered when ΣD ≥ 0.8, and failure is determined when ΣD ≥ 1.0. The remaining useful life is predicted based on the current damage rate: RUL = (1-ΣD) * 100%.

[0110] The above design takes into account individual differences caused by the specific IGBT production process and the impact of actual operating conditions. Using readily available electrothermal parameters that characterize significant degradation, an online IGBT aging prediction method combining analytical physics models with deep learning is proposed, accurately determining the aging status of individual IGBT modules.

[0111] The present invention improves on the classic rainflow algorithm by using online rainflow counting and stress statistics to achieve online IGBT remaining life estimation and evaluation. This eliminates the need for offline module life monitoring, significantly reducing operation and maintenance costs.

[0112] The embodiment of the present invention not only considers temperature factors, but also incorporates other aging effects such as current stress and voltage stress. It can more accurately reflect the aging mechanism of the electric-thermal multi-physical field coupling in actual working conditions, and the prediction accuracy will be significantly improved.

[0113] Based on the same inventive concept, the present invention also provides a computer device, which includes: one or more processors and a memory for storing one or more computer programs; the program includes program instructions, and the processor is used to execute the program instructions stored in the memory. The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, which is used to implement one or more instructions, specifically for loading and executing one or more instructions in a computer storage medium to implement the above method.

[0114] It should be further explained that, based on the same inventive concept, the present invention also provides a computer storage medium having a computer program stored thereon, which, when executed by a processor, performs the above-described method. The storage medium may be any combination of one or more computer-readable media. The computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: an electrical connection having one or more conductors, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present invention, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0115] Throughout this specification, references to terms such as "one embodiment," "example," or "specific example" indicate that a specific feature, structure, material, or characteristic described in conjunction with that embodiment or example is included in at least one embodiment or example of the present disclosure. In this specification, schematic representations of these terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

[0116] The above shows and describes the basic principles, main features and advantages of the present disclosure. Those skilled in the art should understand that the present disclosure is not limited to the above embodiments. The above embodiments and descriptions are merely illustrative of the principles of the present disclosure. Various changes and improvements may be made to the present disclosure without departing from the spirit and scope of the present disclosure, and such changes and improvements shall fall within the scope of the present disclosure.

[0117] The present invention is not limited to the above-mentioned optimal implementation mode. Under the inspiration of the present invention, anyone can derive various other forms of an online prediction method for welding IGBT aging that integrates analytical physical models and deep learning. All equal changes and modifications made within the scope of the patent application of the present invention should fall within the scope of the present invention.

Claims

1. An online prediction method for welding IGBT aging that integrates analytical physical models and deep learning, characterized by: The following steps are involved: Real-time acquisition of temperature-sensitive electrical parameters of the IGBT module, including saturation voltage drop, collector current, case temperature, current stress, and voltage stress; Inputting the temperature-sensitive electrical parameters into a pre-trained deep neural network model to output the junction temperature of the IGBT; Adopting the dynamically updated online rain flow counting method to process the junction temperature fluctuation data, and to count the junction temperature fluctuation amount within the equivalent cycle in real time; Inputting the junction temperature fluctuation into a life prediction model based on the Coffin-Manson equation to calculate the number of failure cycles; Accumulate damage based on linear cumulative damage theory; When the damage amount reaches a first threshold, an early warning is triggered; when it reaches a second threshold, failure is determined, and the remaining service life is predicted based on the current damage amount.

2. The online prediction method for welding IGBT aging by integrating analytical physical model and deep learning according to claim 1 is characterized by: The online rain flow counting method dynamically updates the temperature extreme value through the dual buffer area and outputs the junction temperature fluctuation magnitude in real time.

3. The online prediction method for welded IGBT aging by integrating analytical physical model and deep learning according to claim 1 is characterized by: The parameters of the life prediction model are updated every 5000 cycles.

4. The online prediction method for welded IGBT aging by integrating analytical physical model and deep learning according to claim 1 is characterized by: The deep neural network model includes 3 hidden layers.

5. The online prediction method for welding IGBT aging by integrating analytical physical model and deep learning according to claim 1 is characterized by: The current stress and voltage stress are collected in real time by a Hall sensor.

6. The online prediction method for welding IGBT aging by integrating analytical physical model and deep learning according to claim 1 is characterized by: The remaining useful life is calculated using the following formula: RUL=(1-ΣD)×100%; where ΣD is the accumulated damage.

7. The online prediction method for welding IGBT aging by integrating analytical physical model and deep learning according to claim 1 is characterized by: The first threshold is 0.8, and the second threshold is 1.

0.

8. An online prediction system for welding IGBT aging that integrates analytical physical models and deep learning, characterized by: include: Data acquisition module: real-time acquisition of IGBT temperature-sensitive electrical parameters including saturation voltage drop, collector current, case temperature, current stress, and voltage stress; Junction temperature monitoring module: a pre-trained deep neural network model that inputs the temperature-sensitive electrical parameters and outputs the junction temperature; Dynamic processing module: uses online rain flow counting method to process junction temperature fluctuation data and outputs junction temperature fluctuation quantity in real time; Life prediction module: calculates the number of failure cycles based on the Coffin-Manson equation; Damage accumulation module: accumulates damage and triggers warning / failure; Output module: predicts the remaining service life; the remaining service life is calculated using the following formula: RUL=(1-ΣD)×100%; where ΣD is the accumulated damage amount.

9. A computer device comprising a processor and a memory, wherein the memory stores a computer program, wherein: When the program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images