An online monitoring method for IGBT chip solder layer failure

By establishing a three-dimensional simulation model of the IGBT module and electromagnetic field-temperature field coupling calculation, combining BP neural network and MATLAB fitting neural network, online monitoring of IGBT chip solder layer faults is realized, and fault evaluation in opaque and complex environments of the package structure is solved The problem is to accurately classify and evaluate the welding layer faults, ensuring the health status monitoring and failure prevention of IGBT modules.

CN115291074BActive Publication Date: 2024-11-26HEFEI UNIV OF TECH +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202210973862.7
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-08-15
Publication Date
2024-11-26
Estimated Expiration
2042-08-15

AI Technical Summary

Technical Problem

The packaging structure of the IGBT module is opaque and the working environment is complex, which makes it difficult to classify and evaluate the degree without affecting normal operation, especially the problems of hollowing, cracks, and shedding of the solder layer due to different thermal expansion coefficients and process defects. Impacts its working reliability and lifespan.

Method used

Establish a three-dimensional simulation model of the IGBT module, obtain health factors through fault simulation and electromagnetic field-temperature field coupling calculation, establish a fault classification diagnosis model and fault severity evaluation model, and use BP neural network and MATLAB fitting neural network for online monitoring and real-time acquisition Health factors are evaluated for failure type and severity.

Benefits of technology

The online classification and severity assessment of IGBT chip solder layer faults have been realized, and the types and degrees of failures such as holes, cracks or falls can be accurately identified, and the modules with serious failures can be replaced in a timely manner to avoid major losses caused by failure shutdown.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN115291074B_ABST
    Figure CN115291074B_ABST
Patent Text Reader

Abstract

An online monitoring method for IGBT chip solder layer faults, firstly establishes an IGBT module three-dimensional simulation model, then simulates the fault of the three-dimensional simulation model according to the fault type and severity of the chip solder layer, and then obtains the health factor of the IGBT module through simulation calculation. The health factor includes the maximum temperature, minimum temperature, and voltage drop difference between the collector and the emitter of the IGBT module, and summarizes the health factors obtained under different fault types and fault severity to establish a three-dimensional data set of health factor-fault type-fault severity, establishes a fault classification diagnosis model through a classification algorithm, and establishes a fault severity assessment model through a fitting algorithm, and finally substitutes the measured IGBT module health factor to obtain the actual fault type and fault severity. This method can classify the fault of the IGBT chip solder layer and assess the fault severity.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the technical field of semiconductor devices, and in particular relates to an online monitoring method for an IGBT chip solder layer fault. Background Art

[0002] Insulated Gate Bipolar Transistor (IGBT) has become a representative device in the field of modern power electronics due to its advantages such as high input impedance, simple control circuit, high current density, fast switching speed, and low saturation voltage drop. Due to the continuous increase in the current density and voltage level of IGBT modules and the different thermal expansion coefficients of the various layers of materials in the IGBT modules, coupled with inevitable process defects, the solder layer of the IGBT module is prone to thermal problems such as voids, cracks, and shedding to varying degrees after long-term operation, which seriously affects the working reliability and overall life of the IGBT.

[0003] The difficulty in classifying and diagnosing the fault severity of the IGBT chip solder layer lies in the fact that the IGBT module is an opaque package structure and the working environment of the IGBT is mostly complex, which does not allow for arbitrary suspension of work and disassembly inspection. Therefore, it is particularly important to classify the faults of the IGBT solder layer and evaluate the fault severity without affecting the normal operation of the IGBT and without disassembling the IGBT packaging structure. Summary of the invention

[0004] The purpose of the present invention is to overcome the above problems existing in the prior art and to provide an online monitoring method for IGBT chip solder layer faults which can realize fault classification and fault degree evaluation.

[0005] To achieve the above objectives, the present invention provides the following technical solutions:

[0006] An online monitoring method for an IGBT chip solder layer fault, the online monitoring method is performed in the following steps in sequence:

[0007] S1. Establish a three-dimensional simulation model of the IGBT module;

[0008] S2, firstly perform fault simulation on the three-dimensional simulation model according to the fault type and fault severity of the chip welding layer, then perform conditional loading on the three-dimensional simulation model after the fault simulation, and obtain the health factor of the IGBT module through electromagnetic field-temperature field coupling calculation;

[0009] The fault type is voids, cracks or shedding of the chip soldering layer, the fault severity is void rate, crack rate or shedding degree, and the health factor includes the maximum temperature, minimum temperature, and voltage drop difference between the collector and emitter of the IGBT module;

[0010] S3, first summarize the health factors of the chip welding layer under different fault types and fault severity, and establish a three-dimensional data set of health factor-fault type-fault severity, and then establish a fault classification diagnosis model through a classification algorithm based on the three-dimensional data set, and establish a fault severity assessment model through a fitting algorithm;

[0011] S4. Substitute the IGBT module health factor obtained by real-time acquisition into the fault classification diagnosis model and the fault severity assessment model to obtain the actual fault type and actual fault severity of the IGBT module solder layer.

[0012] In step S3, in the classification algorithm, the health factor in the three-dimensional data set is used as input data and the fault type is used as output data to establish a fault classification diagnosis model for the IGBT chip welding layer. In the fitting algorithm, the health factor in the three-dimensional data set is used as input data and the fault severity is used as output data to perform data fitting to establish a fault severity assessment model for the IGBT chip welding layer.

[0013] In step S2, the electromagnetic field-temperature field coupling calculation step is specifically as follows: firstly, the temperature rise distribution and the potential distribution when the IGBT module is working stably are calculated, and then the maximum temperature, the minimum temperature, and the voltage drop difference between the collector and the emitter of the IGBT module are obtained according to the temperature rise distribution and the potential distribution, wherein the temperature rise distribution and the potential distribution are calculated according to the following formula:

[0014]

[0015]

[0016]

[0017]

[0018]

[0019]

[0020] q 0 =h(T ext -T);

[0021] In the above formula, is the divergence operator, is the current density, Q j,v is the free charge, σ is the dielectric constant, is the electric field, j is the current density, ω is the angular velocity, is the electric displacement field, is the external current, V is the potential distribution, is the heat flux, k is the thermal conductivity, is the temperature distribution, ρ is the density, C p is the heat capacity at constant pressure, is the velocity field defined by the translation subnode when the 3D simulation model moves in the material frame, Q is the heat flow, n is the normal vector, and q 0 is the boundary heat flux, h is the heat transfer coefficient, T ext is the internal temperature and T is the external temperature.

[0022] In step S2, a cylindrical cavity is firstly opened in the middle of the chip solder layer of the three-dimensional simulation model to simulate the occurrence of a cavity failure in the chip solder layer, and then the height of the cylindrical cavity is made equal to the thickness of the chip solder layer, and the radius of the cylindrical cavity is gradually increased to gradually increase the cavity rate, so as to simulate the gradual increase in the degree of cavity failure, until the maximum temperature of the device in the three-dimensional simulation model exceeds the maximum specified temperature. The expression of the cavity rate is:

[0023] Void ratio = upper surface area of ​​cylindrical voids / upper surface area of ​​the chip solder layer without voids.

[0024] In step S2, a rectangular cavity is first opened in the middle of the three-dimensional simulation model to simulate a crack failure in the chip solder layer, and then the width of the rectangular cavity is made equal to the width of the chip solder layer and the height is 0.1-0.3 times the thickness of the chip solder layer. The length of the rectangular cavity is gradually increased to gradually increase the crack rate to simulate the gradual increase in the degree of crack failure until the maximum temperature of the device in the three-dimensional simulation model exceeds the maximum specified temperature. The expression of the crack rate is:

[0025] Crack rate = length of rectangular cavity / length of chip solder layer without cracks.

[0026] In step S2, a cylinder is first set in the middle of the chip solder layer of the three-dimensional simulation model, and the cylindrical part is used as the remaining part of the chip solder layer after the chip solder layer fails to fall off. Then, the height of the cylinder is equal to the thickness of the chip solder layer, and the radius of the cylinder is gradually reduced to gradually increase the degree of falling off, so as to simulate the gradual increase in the degree of falling off failure, until the maximum temperature of the device in the three-dimensional simulation model exceeds the maximum specified temperature. The expression of the falling off degree is:

[0027] Degree of shedding = (upper surface area of ​​the chip solder layer when it is not shedding - upper surface area of ​​the remaining shedding portion) / upper surface area of ​​the chip solder layer when it is not shedding.

[0028] The classification algorithm is a BP neural network. The process of establishing a fault classification diagnosis model using the BP neural network is as follows: first, a three-layer neural network topology model is built, in which the input layer, hidden layer, and output layer in the three-layer neural network topology model are each set to one layer, the number of output layer nodes is set to 3, the number of output layer nodes is set to 1, and the number of hidden layer nodes is set to 3. Then, the three-dimensional data set is divided into a training set and a test set. Through data training and testing and continuous adjustment of the training set and test set division ratio, a BP neural network model with the highest diagnostic accuracy is obtained. This BP neural network model is the fault classification diagnosis model.

[0029] The fitting algorithm is the FITNET function fitting neural network in MATLAB. The process of using the FITNET function fitting neural network to establish a fault severity assessment model is as follows: first, a four-layer neural network topology model is built, in which the input layer, the hidden layer, and the output layer are set to one layer, two layers, and one layer respectively, and the number of nodes in the two hidden layers is 20 and 10 respectively. Then, the data in the three-dimensional data set are randomly arranged and divided into a training set and a test set in a ratio of 8:2. The FITNET function fitting neural network model is obtained through data training and testing. The FITNET function fitting neural network model is the fault severity assessment model.

[0030] The step 1 specifically comprises: first simplifying the IGBT module, and then establishing a three-dimensional simulation model of the IGBT module in COMSOL Multiphysics simulation software according to the size and manufacturing material of the simplified IGBT module.

[0031] The simplified steps are: selecting an IGBT unit in the IGBT module as a simulation modeling object, and ignoring the anti-parallel diode with extremely small heat dissipation, the IGBT module gate, the driver current passing through the gate, and the gate bonding wire during the simulation modeling.

[0032] Compared with the prior art, the present invention has the following beneficial effects:

[0033] In an online monitoring method for an IGBT chip solder layer fault of the present invention, a three-dimensional simulation model of an IGBT module is first established, and then fault simulation is performed on the three-dimensional simulation model according to the fault type and fault severity of the chip solder layer, and conditional loading is performed on the three-dimensional simulation model after the fault simulation, and the health factor of the IGBT module is obtained by electromagnetic field-temperature field coupling calculation, and then the health factors of the chip solder layer under different fault types and fault severity are summarized, and a health factor-fault type-fault severity three-dimensional data set is established, and a fault classification diagnosis model is established according to the three-dimensional data set through a classification algorithm, and a fault severity assessment model is established through a fitting algorithm, and the IGBT module health factor obtained by real-time collection is substituted into the fault classification diagnosis model and the fault severity assessment model to obtain the actual fault type and actual fault severity of the IGBT module solder layer, the fault type is a cavity, crack or detachment of the chip solder layer, and the fault The severity is the void rate, crack rate or shedding degree, and the health factor includes the maximum temperature, minimum temperature, and voltage drop difference between the collector and emitter of the IGBT module. This method targets the chip welding layer fatigue failure that is very easy to occur in the IGBT module due to thermal problems. First, the health factor of the chip welding layer in the IGBT module when different types and degrees of failure occur is obtained through electrothermal coupling finite element data analysis. Then, a classification algorithm is used to analyze the mapping relationship between different fault types of the chip welding layer and the health factor, and a fitting algorithm is used to analyze the mapping relationship between different fault degrees of the chip welding layer and the health factor. Finally, the actually collected health factor is substituted into the mapping relationship to achieve the purpose of fault classification of the IGBT chip welding layer fault and evaluation of the degree of chip welding layer failure, thereby realizing online monitoring of the health status of the IGBT module, timely replacing the IGBT module with serious faults, and avoiding major losses to the entire system caused by shutdown due to IGBT module failure. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0035] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0036] See also Figure 1 , the specific implementation steps of an online monitoring method for IGBT chip solder layer failure are as follows:

[0037] S1. Establish a 3D simulation model of the IGBT module

[0038] An IGBT unit in the IGBT module (model 5SNA0800N330100) commonly used in converter valves was selected as the simulation modeling object. According to the size and manufacturing materials of the IGBT module provided by ABB, Switzerland, a three-dimensional simulation model of the IGBT module was established in COMSOL Multiphysics simulation software. The anti-parallel diode with very little heat dissipation, the gate of the IGBT module, the gate flow of the driver, and the gate bonding wire were ignored during the simulation modeling.

[0039] S2. Fault simulation

[0040] Performing fault simulation on the three-dimensional simulation model obtained in step S1 according to the fault type and fault severity of the chip soldering layer, wherein the fault type is void, crack or fall-off;

[0041] The specific process of simulating the void fault is as follows: a cylindrical void is dug out in the middle of the chip solder layer of the three-dimensional simulation model, the height of the cylindrical void is equal to the thickness of the chip solder layer (the thickness of the chip solder layer of this model of IGBT module is 0.1 mm), which is used to simulate the void fault in the chip solder layer, and the radius of the cylindrical void is gradually increased to simulate the gradual severity of the void fault until the maximum temperature of the device in the three-dimensional simulation model exceeds the maximum safe operating temperature (the maximum safe operating temperature of this model of IGBT module is 150°C). The expression of the void rate is: void rate = upper surface area of ​​the cylindrical void / upper surface area of ​​the chip solder layer when no void occurs. The larger the void rate, the more serious the void fault.

[0042] The specific process of simulating crack failure is as follows: a rectangular hole is dug out in the middle of the three-dimensional simulation model to simulate a crack failure in the chip solder layer. The width of the rectangular hole is equal to the width of the chip solder layer (the width of the chip solder layer of this model of IGBT module is 14 mm), and the height is 0.2 times the thickness of the chip solder layer. The length of the rectangular hole is gradually increased to simulate the gradual severity of the crack failure until the maximum temperature of the device in the three-dimensional simulation model exceeds the maximum safe operating temperature. The expression of the crack rate is: crack rate = length of the rectangular hole / length of the chip solder layer when no cracks appear. The larger the crack rate, the more serious the crack failure.

[0043] The specific process of simulating the shedding failure is as follows: a cylinder is set in the middle of the chip solder layer of the three-dimensional simulation model, the height of the cylinder is equal to the thickness of the chip solder layer, and is used to simulate the remaining shedding part of the chip solder layer after the shedding failure occurs. The radius of the cylinder is gradually reduced to simulate the gradual severity of the shedding failure until the maximum temperature of the device in the three-dimensional simulation model exceeds the maximum specified temperature. The expression of the shedding degree is: shedding degree = (upper surface area of ​​the chip solder layer when no shedding occurs - upper surface area of ​​the remaining shedding part) / upper surface area of ​​the chip solder layer when no shedding occurs. The larger the shedding degree, the more serious the shedding failure.

[0044] S3. Simulate and calculate the health factor of the chip welding layer under different fault types and fault degrees

[0045] First, the three-dimensional simulation model after fault simulation is loaded with simulation conditions according to the actual working conditions, and then the health factor of the IGBT module is calculated by electromagnetic field-temperature field coupling. The health factor includes the maximum temperature, minimum temperature, and voltage drop difference between the collector and the emitter when the IGBT module is working stably. The electromagnetic field-temperature field coupling calculation step is specifically as follows: first, the temperature rise distribution and potential distribution when the IGBT module is working stably are calculated, and then the maximum temperature, minimum temperature, and voltage drop difference between the collector and the emitter of the IGBT module are obtained according to the temperature rise distribution and the potential distribution. The temperature rise distribution and the potential distribution are calculated according to the following electrothermal coupling formula:

[0046]

[0047]

[0048]

[0049]

[0050]

[0051]

[0052] q 0 =h(T ext -T);

[0053] In the above formula, is the divergence operator, is the current density, Q j,v is the free charge, σ is the dielectric constant, is the electric field, j is the current density, ω is the angular velocity, is the electric displacement field, is the external current, V is the potential distribution, is the heat flux, k is the thermal conductivity, is the temperature distribution, ρ is the density, C p is the heat capacity at constant pressure, is the velocity field defined by the translation subnode when the 3D simulation model moves in the material frame, Q is the heat flow, n is the normal vector, and q 0 is the boundary heat flux, h is the heat transfer coefficient, T ext is the internal temperature, T is the external temperature;

[0054] S4. First, summarize the health factors of the chip welding layer under different fault types and fault degrees, and establish a three-dimensional data set of health factor-fault type-fault severity. Then, in the BP neural network, the health factors in the three-dimensional data set are used as input data and the fault type is used as output data to establish a fault classification diagnosis model for the IGBT chip welding layer. In the FITNET function fitting neural network of MATLAB, the health factors in the three-dimensional data set are used as input data and the fault severity is used as output data to perform data fitting, and a fault severity assessment model for the IGBT chip welding layer is established.

[0055] The specific process of using BP neural network to establish the fault classification diagnosis model is as follows: first, a three-layer neural network topology model is built, in which the input layer, hidden layer, and output layer in the three-layer neural network topology model are each set to one layer, the number of output layer nodes is set to 3, the number of output layer nodes is set to 1, and the number of hidden layer nodes is set to 3, and then the three-dimensional data set is divided into a training set and a test set, and the BP neural network model with the highest diagnosis accuracy is obtained through data training and testing and continuous adjustment of the division ratio of the training set and the test set. The BP neural network model is the fault classification diagnosis model;

[0056] The specific process of using the FITNET function fitting neural network to establish a fault severity assessment model is as follows: first, a four-layer neural network topology model is built, in which the input layer, hidden layer, and output layer are set to one layer, two layers, and one layer respectively, and the number of nodes in the two hidden layers is 20 and 10 respectively, and then the data in the three-dimensional data set are randomly arranged and divided into a training set and a test set in a ratio of 8:2, and a FITNET function fitting neural network model is obtained through data training and testing, and the FITNET function fitting neural network model is the fault severity assessment model;

[0057] S5. Randomly select 34 IGBT modules of the same model as test cases. First, use the temperature sensor to collect the maximum temperature and minimum temperature of the test case when it is working stably in real time. Use real-time monitoring to obtain the voltage drop difference between the collector and the emitter of the test case. Use the maximum temperature, minimum temperature, and voltage drop difference between the collector and the emitter as health factors of the 34 test cases. Then, substitute the health factors of the 34 test cases into the fault classification diagnosis model and fault severity assessment model obtained in step S4 for evaluation. The evaluation results show that 15 test cases have chip welding layer faults. The evaluation results are shown in Table 1. The remaining 19 test cases are presumed to have other faults or are in a healthy state:

[0058] Table 1 Chip soldering layer fault types and fault severity in 15 test cases

[0059]

[0060]

[0061] It can be seen from Table 1 that the online monitoring method for IGBT chip solder layer failure of the present invention can accurately classify the failure type of the IGBT module chip solder layer and evaluate the failure degree of the chip solder layer.

Claims

1. An online monitoring method for IGBT chip solder layer failure, characterized in that: The online monitoring method is carried out in the following steps: S1. Establish a three-dimensional simulation model of the IGBT module; S2, firstly perform fault simulation on the three-dimensional simulation model according to the fault type and fault severity of the chip welding layer, then perform conditional loading on the three-dimensional simulation model after the fault simulation, and obtain the health factor of the IGBT module through electromagnetic field-temperature field coupling calculation; The fault type is voids, cracks or shedding of the chip soldering layer, the fault severity is void rate, crack rate or shedding degree, and the health factor includes the maximum temperature, minimum temperature, and voltage drop difference between the collector and emitter of the IGBT module; S3, first summarize the health factors of the chip welding layer under different fault types and fault severity, and establish a three-dimensional data set of health factor-fault type-fault severity, and then establish a fault classification diagnosis model through a classification algorithm based on the three-dimensional data set, and establish a fault severity assessment model through a fitting algorithm; S4. Substitute the IGBT module health factor obtained by real-time acquisition into the fault classification diagnosis model and the fault severity assessment model to obtain the actual fault type and actual fault severity of the IGBT module solder layer.

2. The online monitoring method for IGBT chip solder layer failure according to claim 1, characterized in that: In step S3, in the classification algorithm, the health factor in the three-dimensional data set is used as input data and the fault type is used as output data to establish a fault classification diagnosis model for the IGBT chip welding layer. In the fitting algorithm, the health factor in the three-dimensional data set is used as input data and the fault severity is used as output data to perform data fitting to establish a fault severity assessment model for the IGBT chip welding layer.

3. An online monitoring method for IGBT chip solder layer failure according to claim 1 or 2, characterized in that: In step S2, the electromagnetic field-temperature field coupling calculation step is specifically as follows: firstly, the temperature rise distribution and the potential distribution when the IGBT module is working stably are calculated, and then the maximum temperature, the minimum temperature, and the voltage drop difference between the collector and the emitter of the IGBT module are obtained according to the temperature rise distribution and the potential distribution, wherein the temperature rise distribution and the potential distribution are calculated according to the following formula: q0=h(T ext -T); In the above formula, is the divergence operator, is the current density, Q j,v is the free charge, σ is the dielectric constant, is the electric field, j is the current density, ω is the angular velocity, is the electric displacement field, is the external current, V is the potential distribution, is the heat flux, k is the thermal conductivity, is the temperature distribution, ρ is the density, C p is the heat capacity at constant pressure, is the velocity field defined by the translation subnode when the 3D simulation model moves in the material frame, Q is the heat flow, n is the normal vector, q0 is the boundary heat flux, h is the heat transfer coefficient, T ext is the internal temperature and T is the external temperature.

4. The online monitoring method for IGBT chip solder layer failure according to claim 1 or 2, characterized in that: In step S2, a cylindrical cavity is firstly opened in the middle of the chip solder layer of the three-dimensional simulation model to simulate the occurrence of a cavity failure in the chip solder layer, and then the height of the cylindrical cavity is made equal to the thickness of the chip solder layer, and the radius of the cylindrical cavity is gradually increased to gradually increase the cavity rate, so as to simulate the gradual increase in the degree of cavity failure, until the maximum temperature of the device in the three-dimensional simulation model exceeds the maximum specified temperature. The expression of the cavity rate is: Void ratio = upper surface area of ​​cylindrical voids / upper surface area of ​​the chip solder layer without voids.

5. An online monitoring method for IGBT chip solder layer failure according to claim 1 or 2, characterized in that: In step S2, a rectangular cavity is first opened in the middle of the three-dimensional simulation model to simulate a crack failure in the chip solder layer, and then the width of the rectangular cavity is made equal to the width of the chip solder layer and the height is 0.1-0.3 times the thickness of the chip solder layer. The length of the rectangular cavity is gradually increased to gradually increase the crack rate to simulate the gradual increase in the degree of crack failure until the maximum temperature of the device in the three-dimensional simulation model exceeds the maximum specified temperature. The expression of the crack rate is: Crack rate = length of rectangular cavity / length of chip solder layer without cracks.

6. An online monitoring method for IGBT chip solder layer failure according to claim 1 or 2, characterized in that: In step S2, a cylinder is first set in the middle of the chip solder layer of the three-dimensional simulation model, and the cylindrical part is used as the remaining part of the chip solder layer after the chip solder layer fails to fall off. Then, the height of the cylinder is equal to the thickness of the chip solder layer, and the radius of the cylinder is gradually reduced to gradually increase the degree of falling off, so as to simulate the gradual increase in the degree of falling off failure, until the maximum temperature of the device in the three-dimensional simulation model exceeds the maximum specified temperature. The expression of the falling off degree is: Degree of shedding = (upper surface area of ​​the chip solder layer when it is not shedding - upper surface area of ​​the remaining shedding portion) / upper surface area of ​​the chip solder layer when it is not shedding.

7. The online monitoring method for IGBT chip solder layer failure according to claim 2, characterized in that: The classification algorithm is a BP neural network. The process of establishing a fault classification diagnosis model using the BP neural network is as follows: first, a three-layer neural network topology model is built, in which the input layer, hidden layer, and output layer in the three-layer neural network topology model are each set to one layer, the number of output layer nodes is set to 3, the number of output layer nodes is set to 1, and the number of hidden layer nodes is set to 3. Then, the three-dimensional data set is divided into a training set and a test set. Through data training and testing and continuous adjustment of the training set and test set division ratio, a BP neural network model with the highest diagnostic accuracy is obtained. This BP neural network model is the fault classification diagnosis model.

8. The online monitoring method for IGBT chip solder layer failure according to claim 2, characterized in that: The fitting algorithm is the FITNET function fitting neural network in MATLAB. The process of using the FITNET function fitting neural network to establish a fault severity assessment model is as follows: first, a four-layer neural network topology model is built, in which the input layer, the hidden layer, and the output layer are set to one layer, two layers, and one layer respectively, and the number of nodes in the two hidden layers is 20 and 10 respectively. Then, the data in the three-dimensional data set are randomly arranged and divided into a training set and a test set in a ratio of 8:

2. The FITNET function fitting neural network model is obtained through data training and testing. The FITNET function fitting neural network model is the fault severity assessment model.

9. An online monitoring method for IGBT chip solder layer failure according to claim 1 or 2, characterized in that: Specifically, S1 is as follows: first simplifying the IGBT module, and then establishing a three-dimensional simulation model of the IGBT module in COMSOL Multiphysics simulation software according to the size and manufacturing material of the simplified IGBT module.

10. The online monitoring method for IGBT chip solder layer failure according to claim 9, characterized in that: The simplified steps are: selecting an IGBT unit in the IGBT module as a simulation modeling object, and ignoring the anti-parallel diode with extremely small heat dissipation, the IGBT module gate, the driver current passing through the gate, and the gate bonding wire during the simulation modeling.