Method for estimating cause of partial discharge of power semiconductor module, and device for estimating cause of partial discharge of power semiconductor module

By applying test voltage of voltage mode change in power semiconductor module, measuring the charge amount and using the characteristic amount to infer the cause of local discharge, the problem of difficulty in distinguishing void changes in the prior art is solved, and the durability and reliability of the module are improved .

CN118974572APending Publication Date: 2024-11-15HITACHI POWER SEMICON DEVICE LTD
View PDF 1 Cites 0 Cited by

Patent Information

Application Number
CN202280093157.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Priority Date
2022-03-11
Filing Date
2022-11-08
Publication Date
2024-11-15

AI Technical Summary

Technical Problem

The prior art is difficult to effectively distinguish the void changes in the soft and hard resin parts of the power semiconductor module through local discharge tests, making it difficult to determine the cause of local discharge, affecting the durability and reliability of the module.

Method used

By applying the test voltage mode of the voltage change, the charge amount caused by local discharge is measured, the average value of the charge amount is extracted as the characteristic quantity, and the cause of local discharge is inferred by these characteristic quantity, and the machine learning classification method is combined with the machine learning classification method to automatically distinguish the time-dependent and non-dependent in the type of space of time.

Benefits of technology

It realizes simple and automatic inference of local discharge causes, improves the durability and reliability of power semiconductor modules, and reduces the occurrence of faults and breakdowns.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118974572A_ABST
    Figure CN118974572A_ABST
Patent Text Reader

Abstract

The invention provides a method for estimating the cause of partial discharge of a power semiconductor module, which can easily and automatically estimate the cause of partial discharge by using time series data of a discharge charge amount in a partial discharge test. A method for estimating the cause of partial discharge of a power semiconductor module, which is for estimating the cause of partial discharge during a partial discharge test of the power semiconductor module, comprises: (a) a measurement step for applying a test voltage pattern in which the voltage pattern changes to the power semiconductor module, and measuring the cause of partial discharge of the power semiconductor module; measuring the quantity of electric charges caused by partial discharge of the power semiconductor module; (b) a feature quantity extraction step for extracting a plurality of feature quantities including at least a first feature quantity, which is the average value of the charge quantities in the first period, and a second feature quantity, which is the average value of the charge quantities in the second period; and (c) an estimation step for estimating the cause of partial discharge on the basis of the plurality of feature quantities.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to a test method and a test device for a power semiconductor module, and in particular to a technology that can be effectively applied to a partial discharge test. Background Art

[0002] Power semiconductor modules are widely used in various fields such as electric vehicles, electrified railways, and renewable energy power generation systems, and are required to achieve higher durability and higher reliability.

[0003] One of the causes of failure of power semiconductor modules is insulation breakdown caused by partial discharge. Partial discharge is a weak discharge (corona discharge) that occurs at the defective part of the insulator, which promotes the degradation of the insulator. When partial discharge occurs in the driving voltage, the product life will be shortened, leading to sudden failure and breakdown. Therefore, it is important to detect potential defects that cannot be found by other withstand voltage tests through partial discharge tests, which is very effective for evaluating the insulation reliability of power semiconductor modules.

[0004] As background technology in this technical field, there is, for example, the technology described in Patent Document 1. Patent Document 1 discloses a partial discharge measurement system that can be used to identify the location and cause of partial discharge in diagnosing insulation degradation of a high-voltage device.

[0005] Prior art literature

[0006] Patent Literature

[0007] Patent Document 1: Japanese Patent Application Publication No. 2018-185223 Summary of the invention

[0008] Technical problem to be solved by the invention

[0009] However, it is expected that there are many causes of partial discharge in the power semiconductor module, but disassembly and analysis of the power semiconductor module requires time and labor, so it takes considerable time and cost to determine the cause of partial discharge.

[0010] In addition, the power semiconductor module uses soft resin and / or hard resin as an insulating material. When there are voids (bubbles) in the soft resin, the size of the void is likely to decrease over time, but when there are voids in the hard resin, the size of the void in the hard resin is less dependent on time. According to the principle of partial discharge, the size of the void is correlated with the amount of discharge charge. By measuring the amount of discharge charge during the partial discharge test, the cause of the partial discharge can be determined. However, as described above, depending on the material, when the state of the existing void changes over time, it is difficult to determine the cause of the partial discharge.

[0011] Patent Document 1 does not describe distinguishing between time-dependent voids and time-independent voids by measurement, and thus it is difficult to distinguish, for example, in which of the soft resin portion and the hard resin portion the partial discharge has occurred.

[0012] Therefore, an object of the present invention is to provide a method and device for estimating a cause of partial discharge of a power semiconductor module, which can easily and automatically estimate the cause of partial discharge using time series data of the amount of discharge charge in a partial discharge test.

[0013] Technical solutions to the problem

[0014] In order to solve the above-mentioned problems, the present invention provides a method for inferring the cause of local discharge of a power semiconductor module, which is used to infer the cause of local discharge during a local discharge test of a power semiconductor module, and is characterized in that it includes: (a) a measuring step, applying a test voltage pattern in which a voltage pattern varies to the power semiconductor module, and measuring the amount of charge caused by the local discharge of the power semiconductor module; (b) a feature quantity extraction step, extracting a plurality of feature quantities including at least an average value of the charge quantity in a first period, i.e., a first feature quantity, and an average value of the charge quantity in a second period, i.e., a second feature quantity; and (c) an inference step, inferring the cause of local discharge based on the plurality of feature quantities.

[0015] In addition, the present invention provides a device for inferring the cause of local discharge of a power semiconductor module, which is used to infer the cause of local discharge during a local discharge test of the power semiconductor module, and is characterized in that it includes: a voltage application unit, which applies a test voltage pattern in which a voltage pattern varies to the power semiconductor module; a current measuring unit, which measures the amount of charge caused by the local discharge of the power semiconductor module; a feature quantity calculation unit, which extracts a plurality of feature quantities including at least a first feature quantity which is an average value of the charge quantity in a first period, and a second feature quantity which is an average value of the charge quantity in a second period; and a cause classification unit, which infers the cause of local discharge based on the plurality of feature quantities extracted by the feature quantity calculation unit.

[0016] Effects of the Invention

[0017] According to the present invention, a method for estimating a cause of partial discharge in a power semiconductor module and a device for estimating a cause of partial discharge in a power semiconductor module can be realized, which can easily and automatically estimate the cause of partial discharge using time series data of the amount of discharge charge in a partial discharge test.

[0018] This can contribute to improving the durability and reliability of the power semiconductor module.

[0019] Problems, features, and effects other than those described above will become clear from the following description of the embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 This is a diagram showing the overall structure of a partial discharge cause estimation device for a power semiconductor module according to a first embodiment of the present invention.

[0021] Figure 2A This is a three-dimensional diagram of a power semiconductor module during a partial discharge test.

[0022] Figure 2B It is a diagram showing a typical example of partial discharge test conditions.

[0023] Figure 2C It is a diagram showing an example of a discharge pattern during a partial discharge test.

[0024] Figure 3 It is a diagram showing an example of a discharge pattern during a partial discharge test.

[0025] Figure 4 This is a diagram showing threshold determination in the method for estimating the cause of partial discharge in the power semiconductor module according to the first embodiment of the present invention.

[0026] Figure 5 It is a diagram showing the flow of each process in the feature quantity calculation unit, the learning execution unit, and the cause classification unit of the partial discharge cause estimation device for the power semiconductor module according to the second embodiment of the present invention.

[0027] Figure 6 This is a diagram conceptually showing the cause classification method using machine learning in Example 2 of the present invention.

[0028] Figure 7 This is a diagram showing a GUI of a partial discharge cause estimation device for a power semiconductor module according to a third embodiment of the present invention.

[0029] Figure 8 It is a diagram showing an example of partial discharge in a power semiconductor module. DETAILED DESCRIPTION

[0030] Hereinafter, embodiments of the present invention will be described using the accompanying drawings. In addition, the same reference numerals are given to the same components in each figure, and detailed descriptions of the overlapping parts are omitted.

[0031] Example 1

[0032] First, refer to Figure 8 Partial discharge in a power semiconductor module will be described. Figure 8 FIG. 1 is a diagram showing an example of partial discharge in a typical power semiconductor module 31 .

[0033] The power semiconductor module 31 is as follows Figure 8As shown, a semiconductor chip 81 such as an IGBT (Insulated Gate Bipolar Transistor) and a diode is mounted. The back of the semiconductor chip 81 is bonded to a metal wiring 83 on the surface of an insulating circuit board 84 by solder 82. A bonding wire 93 is connected to the front of the semiconductor chip 81, and is connected to other metal wirings 83 on the front of the insulating circuit board 84 via the bonding wire 93.

[0034] On the other hand, the metal wiring 85 on the back side of the insulating circuit board 84 is bonded to the heat dissipation base plate 87 by solder 86. On the heat dissipation base plate 87, a case 88 and a cover 89 are bonded and fixed by an adhesive 90 so as to cover the semiconductor chip 81, the metal wiring 83, the insulating circuit board 84, and the metal wiring 85. The case 88 is provided with an auxiliary terminal 92.

[0035] The gaps between the heat dissipation base plate 87, the case 88, the cover 89, the semiconductor chip 81, the metal wiring 83, the insulating circuit board 84, and the metal wiring 85 are filled with soft resin 91 to insulate the inside of the power semiconductor module 31. In addition, hard resin 94 is arranged at the end of the metal wiring 83 on the surface of the insulating circuit board 84 where the electric field is particularly likely to concentrate in order to strengthen the insulation.

[0036] Generally speaking, the heat sink 87 is at ground potential, so when the driving voltage is applied to the power semiconductor module 31, the following is generated: Figure 8 The electric field is in the direction shown by the arrow.

[0037] Here, for example, when voids (air bubbles) exist in the hard resin 94 , the soft resin 91 , or the adhesive 90 , partial discharges P1 to P3 may occur.

[0038] Furthermore, when there is a gap or a clearance between the metal wiring 83 and the insulating circuit board 84 , there is a possibility that a partial discharge P4 may occur.

[0039] Furthermore, partial discharge P5 may also occur at the end of the metal wiring 83 where the hard resin 94 is not arranged.

[0040] Next, refer to Figures 1 to 4 A method for estimating a cause of partial discharge in a power semiconductor module and a device for estimating a cause of partial discharge in a power semiconductor module according to a first embodiment of the present invention will be described.

[0041] Figure 1 It is a diagram showing the overall structure of the partial discharge cause estimation device for a power semiconductor module according to the present embodiment.

[0042] like Figure 1As shown, the partial discharge cause estimation device for a power semiconductor module of the present embodiment includes, as main components, a discharge test device 1, a computer server 2, and a communication unit 3 connecting the discharge test device 1 and the computer server 2.

[0043] The discharge test equipment 1 includes a current measuring unit 5 and a voltage applying unit 6. During the partial discharge test, the electronic device 4 as the discharge test object is connected to the current measuring unit 5 and the voltage applying unit 6, and a test voltage is applied to the electronic device 4 from the voltage applying unit 6. The current value (charge amount) flowing in the electronic device 4 is measured by the current measuring unit 5, thereby measuring the charge amount caused by partial discharge. The electronic device 4 corresponds to the power semiconductor module 31.

[0044] The computer server 2 includes a computing unit 7, a storage unit 8, and a user interface unit (GUI) 9. Figure 1 In the figure, the calculation unit 7, the storage unit 8, and the user interface unit (GUI) 9 are installed in the same place as the computer server 2, but they may be installed in different places and connected to each other by a network.

[0045] The calculation unit 7 includes an inspection execution unit 10 , a signal processing unit 11 , a feature quantity calculation unit 12 , a classification model generation unit 13 , a cause classification unit 14 , a result display execution unit 15 , and a labeling execution unit 16 .

[0046] The storage unit 8 has a partial discharge test program 17, a feature value calculation program 18, a marking program 19, a classification model generation program 20, a classification program 21, a result display program 22, a current signal storage unit 23, a feature value storage unit 24, a classification model generation data storage unit 25, a learned classification model storage unit 26 and a classification inference result storage unit 27.

[0047] The user interface unit (GUI) 9 includes an input unit 28 , a computer communication unit 29 , and a display unit 30 .

[0048] use FIG. 2A to FIG. 2C The basic concept of the present invention is explained. Figure 2A It is a perspective view of the power semiconductor module 31 during a partial discharge test. Figure 2B It is a diagram showing a typical example of partial discharge test conditions. Figure 2C It is a diagram showing an example of a discharge pattern during a partial discharge test.

[0049] like Figure 2A As shown, during the partial discharge test, an AC voltage is applied from the voltage applying unit 6 to the power semiconductor module 31 as a test object, and a current value (discharge charge amount) is measured by the current measuring unit 5 .

[0050] As test conditions, for example, Figure 2B The test voltage pattern shown is a test voltage pattern in which the voltage pattern changes with time, and a test voltage is applied to the power semiconductor module 31 . Figure 2B The vertical axis represents the applied voltage AV, and the horizontal axis represents the test time TT.

[0051] Figure 2B In the example, the voltage is increased from the start of the test to a predetermined voltage V1, and then the applied voltage V1 is maintained for a certain period of time ( Figure 2B After that, the voltage is reduced to a predetermined voltage V2, and then the applied voltage V2 is maintained for a certain period of time ( Figure 2B Furthermore, the specified voltage V3 is maintained for a certain period of time ( Figure 2B The duration of the delay is about 5s), and then the voltage is reduced.

[0052] in addition, Figure 2B Although the voltage V2 and the voltage V3 are shown as the same voltage, the voltage values ​​of the voltage V2 and the voltage V3 may be changed. The test conditions (test voltage pattern) differ for each type of the power semiconductor module 31 .

[0053] exist Figure 2C Shown in the use Figure 2B The test results are obtained with the test voltage mode. Figure 2C The vertical axis represents the charge CA, and the horizontal axis represents the test time TT.

[0054] Figure 2C middle, Figure 2B The application period (about 60 seconds) of the voltage V1 in the test voltage mode corresponds to the period t1. In addition, the application period (about 25 seconds) of the voltage V2 and the application period (about 5 seconds) of the voltage V3 correspond to the period t2 and the period t3, respectively.

[0055] like Figure 2C As shown, during period t1 when a relatively high voltage V1 is applied, the charge amount CA gradually increases while repeatedly moving up and down. On the other hand, during period t2 when a voltage V2 lower than the voltage V1 is applied, the charge amount CA gradually decreases while repeatedly moving up and down.

[0056] Thus, by using a test voltage pattern that varies with time ( Figure 2B ) By applying a test voltage to the power semiconductor module 31, it is possible to obtain timing data (discharge pattern) of the discharge charge amount in the partial discharge test.

[0057] The time series data (discharge pattern) of the discharged charge amount is compared with past data stored in the feature storage unit 24 to associate with the cause of partial discharge, thereby estimating the cause of partial discharge from the time series data (discharge pattern).

[0058] Furthermore, by performing the partial discharge test again after a predetermined period of time has elapsed, it is possible to distinguish between a partial discharge cause that depends on time and a partial discharge cause that does not depend on time.

[0059] use Figure 3 An example in which time series data of the amount of discharged electric charge (discharge pattern) is associated with the cause of partial discharge will be described. Figure 3 This is a diagram showing an example of a discharge pattern during a partial discharge test based on past data.

[0060] Figure 3 HV, SV, and EFC respectively represent the hard resin voids, soft resin voids, and electric field concentration areas that are the causes of partial discharge. IM and RM in the figure respectively represent the initial measurement (first measurement) and the re-measurement (second measurement).

[0061] In the case of the hard resin void HV, discharge occurs from the voids present in the hard resin. It has the characteristics that there is no upward trend during voltage application and the void diameter does not depend much on time. A As shown, the difference between the initial measurement of IM and the re-measurement of RM is small.

[0062] In the case of the soft resin void SV, discharge occurs from the voids present in the soft resin. It has the characteristics that the void diameter may increase during voltage application and may decrease over time. B As shown, the difference between the initial measurement of IM and the re-measurement of RM is large.

[0063] In the case of the electric field concentration area EFC, discharge occurs due to local insulation breakdown of the soft resin at the electrode end and other places where the electric field is concentrated. C The characteristic shown is the rise due to electrical tree degradation during voltage application.

[0064] As described above, in the partial discharge of the power semiconductor module 31 , the discharge mechanism differs for each cause, and therefore the time series data (discharge pattern) during the partial discharge test also differs.

[0065] According to the principle of partial discharge, the size of the voids is correlated with the amount of discharged charge. For example, the size of the soft resin voids SV is likely to decrease over time, but the size of the hard resin voids HV is not much dependent on time.

[0066] Therefore, by performing measurement a plurality of times (for example, initial measurement and re-measurement) and comparing characteristic quantities (for example, average values) indicating the magnitude of the charge amount, the hard resin voids HV and the soft resin voids SV can be distinguished.

[0067] use Figure 4 The following describes how to identify the cause of partial discharge by threshold judgment. Figure 4 This is a diagram showing threshold determination in the method for estimating the cause of partial discharge in the power semiconductor module according to the present embodiment.

[0068] First, determination is started in step S1.

[0069] Next, in step S2, the charge amount at the start of the measurement is compared with a predetermined threshold. When the charge amount is below the predetermined threshold and the charge amount increases over time in the same measurement (BT), it is determined that the cause of the discharge is the electric field concentration part EFC. SM in the figure indicates that the charge amount is below the threshold at the start of the measurement. As a method for determining whether the charge amount increases over time in the same measurement, for example, in the same measurement, the average value of the charge amount in the first period, i.e., the first characteristic value, and the average value of the charge amount in the second period after the first period, i.e., the second characteristic value, are used. When the first characteristic value is greater than the second characteristic value by more than a predetermined threshold, it can be determined that the charge amount increases.

[0070] When the charge amount exceeds the predetermined threshold value (AT), the process proceeds to step S3, and the difference between the average charge amount measured initially and the average charge amount measured again is compared with the predetermined threshold value. When the difference between the average charge amount measured initially and the average charge amount measured again exceeds the predetermined threshold value (AT), it is determined that the cause of discharge is the soft resin void SV. D Indicates that the difference in average values ​​exceeds the threshold value. In this case, for example, it is possible to make a judgment by comparing the first characteristic value in the initial measurement IM with the first characteristic value in the re-measurement RM. In addition, it is also possible to make a judgment by comparing the second characteristic value in the initial measurement IM with the second characteristic value in the re-measurement RM. Alternatively, it is also possible to use the prescribed period in the initial measurement IM as the first period, and the prescribed period in the re-measurement RM corresponding to the first period of the initial measurement IM as the second period, and compare the first characteristic value in the initial measurement IM with the second characteristic value in the re-measurement RM. However, in this case, the judgment of step S2 cannot be performed, so it is preferred to extract the first characteristic value and the second characteristic value in the initial measurement IM, and extract the first characteristic value and the second characteristic value in the re-measurement RM. Thus, it is possible to make judgments in both steps S2 and S3.

[0071] When the difference between the average charge amount measured initially and measured again is less than a predetermined threshold value (BT), the process proceeds to step S4 to determine whether there is an auxiliary terminal 92 of the box 88. When it is determined that there is an auxiliary terminal 92 ("Yes"), it is determined that the cause of the discharge is the adhesive void ADV.

[0072] If it is determined that the auxiliary terminal 92 does not exist ("No"), the process proceeds to step S5 to determine whether there is hard resin. If it is determined that there is hard resin ("Yes"), it is determined that the cause of discharge is the hard resin void HV. E Indicates that the difference in average value is below the threshold value.

[0073] When it is determined that the hard resin does not exist ("No"), it is determined that the discharge cause is other causes OV. Other causes OV include, for example, causes other than gaps such as loose bolts.

[0074] According to the method for estimating a partial discharge cause of a power semiconductor module and the device for estimating a partial discharge cause of a power semiconductor module of the present embodiment described above, the cause of partial discharge can be easily and automatically estimated using time series data of the amount of discharge charge in a partial discharge test.

[0075] Example 2

[0076] Reference Figure 5 and Figure 6 A method for estimating a cause of partial discharge in a power semiconductor module and a device for estimating a cause of partial discharge in a power semiconductor module according to a second embodiment of the present invention will be described. In this embodiment, classification of causes of partial discharge using machine learning will be described.

[0077] Figure 5 It is a diagram showing the flow of each process in the feature quantity calculation unit, the learning execution unit, and the cause classification unit of the partial discharge cause estimation device for the power semiconductor module of the present embodiment.

[0078] like Figure 5 As shown, in the feature quantity calculation unit, the feature quantity extraction mode FEM is executed.

[0079] First, in step S10 , time series data of a partial discharge test is extracted from the current signal storage unit 23 for an ID of interest (product batch, etc.).

[0080] Next, in step S11 , an applied voltage curve of the partial discharge test defined in the partial discharge test program 17 is extracted.

[0081] Next, in step S12, a feature value (for example, an average value) of the time series data is calculated for each required region of the applied voltage curve. In this case, the feature value may be calculated for a plurality of tests.

[0082] Finally, in step S13 , the feature amount is associated with the information of the attention ID and then stored in the feature amount storage unit 24 .

[0083] In the learning execution unit, the learning mode LM is executed.

[0084] First, in step S14 , a data set in which the feature amount and the discharge cause are associated is extracted from the learning data storage unit.

[0085] Next, in step S15, preprocessing such as converting categorical variables into dummy variables and standardizing continuous variables is performed.

[0086] Next, in step S16, a model for classifying causes is generated for the preprocessed data set using at least one machine learning algorithm.

[0087] Next, in step S17, the best classification model with the highest classification accuracy of the cause is selected from the generated generation model group. In this case, ensemble learning using a plurality of machine learning algorithms can also be performed.

[0088] Finally, in step S18 , the optimal classification model and parameters related to preprocessing are stored in the learned classification model storage unit 26 .

[0089] In the cause classification unit, the classification mode CM is executed.

[0090] First, in step S19 , the learned model and parameters related to preprocessing are extracted from the learned classification model storage unit 26 .

[0091] Next, in step S20, feature quantity data of the partial discharge test of the requested ID is extracted from the classification data storage unit.

[0092] Next, in step S21, the discharge causes and the probability of belonging to each cause are calculated based on the feature quantity data, the learned (trained) model, and the parameters related to the preprocessing.

[0093] Finally, in step S22 , the discharge causes and the calculation results of the probability of belonging to each cause are associated with the ID and then stored in the classification estimation result storage unit 27 .

[0094] use Figure 6 Explain the basic concepts of the cause classification method using machine learning. Figure 6 This is a diagram conceptually showing the cause classification method using machine learning in this embodiment.

[0095] First, if Figure 6 As shown in the left figure of , feature values ​​are extracted from the data storing the test conditions and the discharge charge amount data. Figure 6 In the example of , it is assumed that the average value of the test voltage V1 is calculated.

[0096] Then, if Figure 6 As shown in the middle figure, a classification model is generated through machine learning based on a learning data set obtained by associating feature quantities with causes. Figure 6CF and EF in the middle figure represent cause and electric field concentration respectively.

[0097] A learning data set obtained by associating feature quantities with causes is obtained from a learning database 32, and machine learning is performed using a classification algorithm 33 to generate a learned model 34. AC-A and AC-B of the learned model 34 represent feature quantity A and feature quantity B, respectively.

[0098] Then, if Figure 6 As shown in the right figure of , the learned classification model 34 is used to infer the cause based on the new discharge pattern 35 whose cause is unknown. As the inference result, ER outputs the cause as electric field concentration EF. The predicted probability PP at this time is as follows Figure 6 As shown in the right figure.

[0099] Example 3

[0100] Reference Figure 7 A method for estimating a partial discharge cause of a power semiconductor module and a device for estimating a partial discharge cause of a power semiconductor module according to a third embodiment of the present invention will be described. In this embodiment, an example of a GUI (Graphical User Interface) for outputting an estimation result ER will be described. Figure 7 It is a diagram showing a GUI of the partial discharge cause estimation device for a power semiconductor module according to the present embodiment.

[0101] like Figure 7 As shown, the GUI displays a display button 101, a variety / batch selection unit 102, an applied voltage curve / charge amount display unit 103, a registration button 104, a cause label input unit 105, a re-learning button 106 and a discharge cause classification result display unit 107, and the user can use the GUI to input the local discharge cause inference device of the power semiconductor module.

[0102] For example, when the cause of partial discharge is to be confirmed, the product Va and lot Lot of interest are selected using the product / lot selection unit 102 , and the display button 101 is pressed. Figure 7 , an example is shown in which "AA" is selected as the variety Va and "B1R9" is selected as the lot Lot.

[0103] As a result, the applied voltage curve / charge amount display section 103 displays the applied voltage curve and the raw data of the discharged charge amount A of the batch B1R9 of the variety AA selected by the variety / batch selection section 102. At the same time (in groups), the classification result B of the discharge cause is displayed in the classification result display section 107 of the discharge cause. Figure 7 , as an example of the classification result B of the discharge cause, the probabilities of the soft resin void SV, the hard resin void HV, and the electric field concentration portion EFC are shown.

[0104] In addition, for example, when the classification of partial discharge causes using machine learning described in Example 2 is to be executed again (re-learned), the user enters the required cause to be registered in the cause label input unit 105, presses the registration button 104 to register, and then presses the re-learning button 106, thereby performing re-learning. Figure 7 , as an input example in the reason tag input unit 105, an example in which the soft resin space SV is input to the reason tag FL is shown.

[0105] In addition, the present invention is not limited to the above-mentioned embodiments, and includes various modified examples. For example, the above-mentioned embodiments are described in detail to explain the present invention in an easy-to-understand manner, and are not limited to all the structures described. In addition, a part of the structure of a certain embodiment can be replaced with the structure of other embodiments, and the structure of other embodiments can be added to the structure of a certain embodiment. In addition, for a part of the structure of each embodiment, other structures can be added, deleted, or replaced.

[0106] Description of Reference Numerals

[0107] 1 ... discharge test equipment, 2 ... computer server, 3 ... communication unit, 4 ... electronic device (discharge test object), 5 ... current measuring unit, 6 ... voltage applying unit, 7 ... calculation unit, 8 ... storage unit, 9 ... user interface unit (GUI), 10 ... inspection execution unit, 11 ... signal processing unit, 12 ... feature quantity calculation unit, 13 ... classification model generation unit, 14 ... cause classification unit, 15 ... result display execution unit, 16 ... marking execution unit, 17 ... partial discharge test program, 18 ... feature quantity calculation program, 19 ... marking program, 20 ... classification model generation program, 21 ... classification program, 22 ... result display program, 23 ... current signal storage unit, 24 ... feature quantity storage unit, 25 ... data storage unit for classification model generation, 26 ... learned classification model storage unit, 27 ... classification inference result storage unit, 28 ... input unit, 29 ... computer communication unit, 30 ... display unit, 31 ... power semiconductor module, 32 ... learning database , 33...Classification algorithm, 34...Learned model, 35...New discharge mode, 81...Semiconductor chip, 82...Solder, 83...Metal wiring, 84...Insulating circuit board, 85...Metal wiring, 86...Solder, 87...Heat dissipation base plate, 88...Box, 89...Lid, 90...Adhesive, 91...Soft resin, 92...Auxiliary terminal, 93...Bond wire, 94...Hard resin, 101...Display button, 102...Product / lot selection unit, 103... …Applied voltage curve / charge amount display section, 104…Register button, 105…Cause label input section, 106…Relearn button, 107…Discharge cause classification result display section, AV…Applied voltage, TT…Test time, CA…Charge amount, HV…Hard resin void, SV…Soft resin void, ADV…Adhesive void, OV…Other causes, EFC…Electric field concentration section, EF…Electric field concentration, IM…Initial measurement, RM…Remeasurement, D A ...the difference is small, D B ... the difference is large, D C ...the rise caused by the deterioration of electrical tree, D D ...the difference in the mean value exceeds the threshold, D E...the difference in average value is below the threshold, AT...exceeds the threshold, BT...below the threshold, SM...below the threshold at the start of measurement, FEM...feature extraction mode, LM...learning mode, CM...classification mode, CF...cause, AC...feature, ER...inference result, PP...prediction probability, Va...variety, Lot...batch, FL...cause label, A...applied voltage curve and discharge charge, B...classification result of discharge cause, P1 to P5...partial discharge, S1...judgment start, S2...judgment of charge at the start of measurement, S3...judgment of the difference in average charge between the first measurement and the remeasurement, S4...judgment of the presence or absence of auxiliary terminal, S5...judgment of the presence or absence of hard resin.

Claims

1. A method for inferring the cause of partial discharge of a power semiconductor module, for inferring the cause of partial discharge during a partial discharge test of a power semiconductor module, characterized in that: include: (a) a measuring step of applying a test voltage pattern in which a voltage pattern varies to a power semiconductor module, and measuring an amount of charge caused by partial discharge of the power semiconductor module; (b) a feature quantity extraction step of extracting a plurality of feature quantities including at least a first feature quantity, which is an average value of the charge quantity in the first period, and a second feature quantity, which is an average value of the charge quantity in the second period; and (c) An inference step of inferring a cause of partial discharge based on the plurality of characteristic quantities.

2. The method for inferring the cause of partial discharge of a power semiconductor module according to claim 1, wherein: In the step (a), a second measurement is performed after a predetermined time has passed since the first measurement, and in the step (b), a feature quantity obtained in the first measurement and a feature quantity obtained in the second measurement are extracted. In the step (c), the cause of partial discharge is estimated using the characteristic amount obtained in the first measurement and the characteristic amount obtained in the second measurement.

3. The method for inferring the cause of partial discharge of a power semiconductor module according to claim 2, wherein: In the step (b), a plurality of feature quantities obtained in the first measurement and a plurality of feature quantities obtained in the second measurement are extracted. In the step (c), the cause of partial discharge is estimated using the characteristic amount obtained in the first measurement and the characteristic amount obtained in the second measurement.

4. The method for inferring the cause of partial discharge of a power semiconductor module according to claim 2, wherein: In the step (c), when the difference between the characteristic amount obtained in the first measurement and the characteristic amount obtained in the second measurement is equal to or larger than a predetermined threshold value, it is estimated that the partial discharge is caused by voids in the soft resin.

5. The method for inferring the cause of partial discharge of a power semiconductor module according to any one of claims 1 to 4, characterized in that: include: (d) a learning data set extraction step, extracting a learning data set obtained by associating the feature quantity with the discharge cause from the learning database; and (e) a cause classification model generation step, for the learning data set, using at least one machine learning algorithm to generate a model for classifying causes, In the step (c), the cause of the partial discharge is inferred using the cause classification model generated in the step (e).

6. The method for inferring the cause of partial discharge of a power semiconductor module according to claim 5, characterized in that: After the step (c), the test voltage pattern is displayed on a GUI (Graphical User Interface) together with the inferred partial discharge cause.

7. The method for inferring the cause of partial discharge of a power semiconductor module according to claim 6, wherein: After the discharge cause requested by the user is registered using the GUI, re-learning is instructed to perform re-learning in the step (e) on the learning data set including the registered discharge cause.

8. A device for inferring the cause of partial discharge of a power semiconductor module, used for inferring the cause of partial discharge during a partial discharge test of a power semiconductor module, characterized in that: include: A voltage applying unit that applies a test voltage pattern in which a voltage pattern varies to the power semiconductor module; a current measuring unit for measuring an amount of charge caused by partial discharge of the power semiconductor module; a feature quantity calculation unit that extracts a plurality of feature quantities including at least a first feature quantity that is an average value of the charge quantity in the first period and a second feature quantity that is an average value of the charge quantity in the second period; and A cause classification unit estimates a cause of partial discharge based on the plurality of feature quantities extracted by the feature quantity calculation unit.

9. The device for inferring the cause of partial discharge of a power semiconductor module according to claim 8, characterized in that: include: A data storage unit for generating a classification model; a classification model generating unit, which extracts a learning data set obtained by associating feature quantities with discharge causes from the classification model generating data storing unit, and generates a model for classifying causes using at least one machine learning algorithm for the learning data set; and The cause classification unit estimates a cause of partial discharge using the classification model generated by the classification model generation unit.

10. The device for inferring the cause of partial discharge of a power semiconductor module according to claim 9, characterized in that: A GUI (Graphical User Interface) is provided for displaying the test voltage pattern together with the inferred cause of partial discharge.

11. The device for inferring the cause of partial discharge of a power semiconductor module according to claim 10, characterized in that: After the discharge cause requested by the user is registered using the GUI, re-learning is instructed, and the classification model generation unit performs re-learning on the learning data set including the registered discharge cause.

Citation Information

Patent Citations

  • Partial discharge measurement system and partial discharge measurement method

    JP2018185223A