Power module temperature hybrid model prediction method

Abstract

The invention discloses a power module temperature hybrid model prediction method, and relates to the technical field of reliability of power electronic devices, and the method comprises the following steps: building a power module finite element thermal model based on the geometric dimension and internal characteristics of a power module; acquiring a temperature field sampling snapshot based on the power module finite element thermal model; extracting a stable evolution trend and an unstable evolution trend of a temperature field of the power module in the sampling snapshot based on a dynamic mode decomposition method; and predicting a stable trend by using an iteration method, and predicting an unstable trend by using a deep learning model. And superposing the predicted stable trend and the predicted unstable trend to obtain a predicted snapshot of the temperature field. According to the method, hybrid model prediction is adopted, the calculation efficiency can be improved as much as possible while the simulation precision is guaranteed, the chip temperature and internal temperature data of the power module are obtained, the number of needed sampling snapshots is reduced, and the modeling period is shortened.

Description

Technical Field

[0001] The present invention relates to the technical field of power electronic device reliability, and in particular to a power module temperature hybrid model prediction method. Background Art

[0002] Power modules, due to their high power density and high efficiency, have become core components in electric vehicle power converters. Research has shown that the lifespan of a power module is closely related to its maximum junction temperature and its temperature fluctuation, which can be quantified using lifetime prediction models. Accurately predicting the temperature field of power modules is crucial to effectively predicting their service life and improving the overall reliability of power electronics systems.

[0003] The finite element method (FEM) is a numerical solution based on materials and structures. It can be used to accurately solve the temperature and stress characteristics of power modules. However, due to the large consumption of computing resources and the long analysis time, the finite element analysis method is difficult to use for transient analysis of power module junction temperature.

[0004] Therefore, a hybrid model prediction method for the power module temperature field is needed to improve the calculation efficiency while ensuring the calculation accuracy. Summary of the Invention

[0005] The problem to be solved by the present invention is to provide a power module temperature hybrid model prediction method, which can greatly improve the calculation efficiency and ensure the calculation accuracy of the power module temperature field fast calculation method.

[0006] The present invention adopts the following technical solution: a power module temperature hybrid model prediction method, comprising the following steps:

[0007] Step 1: Build a three-dimensional finite element thermal model of the power module and define material properties and boundary conditions;

[0008] Step 2: Obtain the junction-to-case thermal resistance response curve of the power module through finite element simulation and compare it with the junction-to-case thermal resistance curve in the data sheet to verify the accuracy of the 3D finite element thermal model.

[0009] Step 3: Obtain sampling snapshots of the power module temperature field through finite element simulation, where the sampling interval is Δt and the number of sampling snapshots is n;

[0010] Step 4: Based on the dynamic mode decomposition method and the deep learning AE-LSTM model, a hybrid prediction model of the temperature field of the power module is established;

[0011] Step 5: Extract the evolution mode of the temperature field by the dynamic mode decomposition method, screen the stable mode and unstable mode according to the frequency and growth rate of the evolution mode, and obtain the stable trend under the stable mode and the unstable trend under the unstable mode;

[0012] Step 6: Use the recursive method to predict future stable trend snapshots, and use the deep learning AE-LSTM model to predict future unstable trend snapshots;

[0013] Step 7: Superimpose the future stable trend snapshot and the future unstable trend snapshot to obtain the future temperature field snapshot.

[0014] Preferably, in step 1, the three-dimensional finite element thermal model of the power module is constructed by the following sub-steps:

[0015] Step 1.1: Based on the internal structural features and geometric dimensions of the power module, use 3D drawing software to create a 3D physical model of the power module in actual proportion.

[0016] Step 1.2: Mesh the 3D physical model of the power module and set the boundary conditions and material properties of the 3D physical model;

[0017] Step 1.3: Set the power module chip as the heat source and set the boundary condition to power loss.

[0018] Preferably, in step 4, the hybrid prediction model of the temperature field of the power module is constructed by the following sub-steps:

[0019] Step 4.1: Obtain a sampling snapshot of the power module temperature field based on finite element simulation and construct the temperature data matrix T as the input of the prediction model:

[0020] T=[T1 T2...T i ...T n ]

[0021] Among them, T i Represents the snapshot data of the i-th temperature field, i = 1, 2, ..., n;

[0022] Step 4.2: Based on the temperature data matrix T, the dynamic mode decomposition method is used to obtain the dynamic mode matrix of the temperature field And the mode coefficient matrix λ:

[0023]

[0024] λ=[λ1,λ2,...,λ r ]

[0025] Among them, λ j 、 denote the jth coefficient and dynamic mode respectively.

[0026] Preferably, in step 5, the frequency f of each temperature field evolution mode is calculated according to the mode coefficient matrix j and growth rate g j , select frequency f j and growth rate g j The mode with all values ​​set to 0 is considered the stable mode.

[0027]

[0028] Based on the sampling snapshot and stable mode, the stable trend matrix is ​​obtained by iteration.

[0029]

[0030] Based on the temperature data matrix T, the unstable trend matrix Can be obtained:

[0031]

[0032] in, represents the kth stable mode, λ sj represents the coefficient corresponding to the jth dominant mode, and T1 represents the initial temperature field snapshot.

[0033] Preferably, in step 6, the future stable trend is predicted using the recursive method Using deep learning AE-LSTM model to predict future unstable trends Superimpose the future stable trend snapshot and the future unstable trend snapshot to obtain the future temperature field snapshot T pred .

[0034] Preferably, the deep learning AE-LSTM model is constructed in the following sub-steps:

[0035] Step 6.1: Unstable trend matrix As the training data of the AE model, the compressed feature matrix X of the unstable trend matrix is ​​obtained using the trained AE model;

[0036] Step 6.2: Use the compressed feature matrix X as the training data for the LSTM model and obtain the trained LSTM model through iterative training.

[0037] Step 6.3: Use the trained LSTM model to predict the future compressed feature matrix X pred ;

[0038] Step 6.4: Use the trained AE model to convert the future compressed feature matrix X predRevert to a snapshot of the future unstable trend forecast

[0039] The technical solution of the present invention further provides: an electronic device, comprising:

[0040] one or more processors;

[0041] a storage device having one or more programs stored thereon;

[0042] When the one or more programs are executed by the one or more processors, the one or more processors implement any of the above-mentioned power module temperature hybrid model prediction methods.

[0043] The technical solution of the present invention also provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the steps in any of the above-mentioned power module temperature hybrid model prediction methods are implemented.

[0044] Compared with the prior art, the present invention adopts the above technical solution and has the following technical effects:

[0045] 1. The hybrid model prediction method for the power module temperature field proposed in the present invention can improve the calculation efficiency as much as possible while ensuring the simulation accuracy by adopting the hybrid model prediction method.

[0046] 2. The hybrid model prediction method for the power module temperature field proposed in the present invention can predict the temperature field of the power module including the temperature inside the module package, rather than just the junction temperature.

[0047] 3. The hybrid model prediction method for the power module temperature field proposed in the present invention can reduce the number of required sampling snapshots and shorten the modeling cycle. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 This is a flowchart of the hybrid model prediction method for power module temperature field of the present invention;

[0049] Figure 2 A schematic diagram of transient temperature response calculation according to an embodiment of the present invention;

[0050] Figure 3 Schematic diagram of temperature field calculation according to an embodiment of the present invention. DETAILED DESCRIPTION

[0051] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.

[0052] Throughout this specification, references to terms such as "one embodiment," "example," or "specific example" indicate that the specific features, structures, materials, or characteristics described in conjunction with that embodiment or example are included in at least one embodiment or example of the present invention. 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.

[0053] In one embodiment of the present invention, see Figure 1 , provides a hybrid model prediction method for power module temperature field.

[0054] In this example, a finite element model of the power module was first constructed using Ansys software, defining material properties and boundary conditions. The finite element model discretizes the power module structure into multiple grid points and, in combination with the material thermal parameters and boundary conditions, solves the heat conduction equation for each grid point.

[0055] Then, through numerical analysis methods, the finite element model can simulate the heat transfer process inside the power module, thereby calculating the transient changes in the power module temperature field. The finite element simulation obtains sampling snapshots of the power module temperature field, with a sampling interval of Δt and a number of sampling snapshots of n.

[0056] Next, a hybrid prediction model for the power module's temperature field was built. The acquired sampling snapshots were used to extract the temperature field evolution pattern using a dynamic modal decomposition model. Stable modes were selected based on the mode frequency and growth rate. Stable and unstable trends were then derived using the stable modes and sampling snapshots.

[0057] Finally, an iterative method is used to predict future stable trends, while an AE-LSTM model is used to predict future unstable trends. The future stable and unstable trends are superimposed to obtain a predicted temperature field snapshot. Mesh data from the power module's finite element model is extracted and combined with the predicted temperature field snapshot to generate a temperature field visualization cloud map.

[0058] The specific implementation steps of the method in this embodiment are as follows:

[0059] Step S1: Build a finite element model of the power module. The sub-steps are as follows:

[0060] (1) Based on the internal structural characteristics and geometric dimensions of the power module, a three-dimensional physical model of the power module is constructed in actual proportion using three-dimensional drawing software;

[0061] (2) Meshing the established three-dimensional model and configuring the model’s boundary conditions and material property parameters in the finite element analysis software Ansys;

[0062] (3) Set the power module chip as the heat source and set the boundary condition to power loss.

[0063] Step S2: Build a power module temperature field prediction model. The sub-steps are as follows:

[0064] (1) In MATLAB software, the temperature data from the 1st to the nth sampling snapshots are constructed into a temperature data matrix T as the input of the prediction model:

[0065] T=[T1 T2 T3...T i ...T n ]

[0066] (2) Using the dynamic mode decomposition method to obtain the dynamic mode matrix of the temperature field And its mode coefficient matrix λ:

[0067]

[0068] λ=[λ1,λ2,...,λ r ]

[0069] Step S3: Extract the evolution mode of the temperature field by the dynamic mode decomposition method, screen the stable mode and the unstable mode according to the frequency and growth rate of the evolution mode, and obtain the stable trend in the stable mode and the unstable trend in the unstable mode. The sub-steps are as follows:

[0070] (1) Obtain the frequency f of the mode according to the mode coefficient j and growth rate g j :

[0071]

[0072] Among them, the Re function represents the real part of the complex number, and the Im function represents the imaginary part of the complex number.

[0073] (2) Select frequency f j and growth rate g j The mode of 0 is regarded as stable mode:

[0074] (3) Using iterative methods to obtain stable trends based on stable patterns and sampling snapshots and unstable trends

[0075]

[0076]

[0077] Step S4: Use recursion method to predict future stable trends Using AE-LSTM model to predict unstable trends

[0078]

[0079] Step S5: superimpose the future stable trend snapshot and the future unstable trend snapshot to obtain the future temperature field snapshot T pred .

[0080]

[0081] In particular, in this embodiment, the steps for building the AE-LSTM model are as follows:

[0082] (1) Unstable trend matrix As the training data of the AE model, the trained AE model is used to obtain the compressed feature matrix X of the unstable trend matrix.

[0083] (2) The unstable trend compression feature matrix is ​​used as the training data of the LSTM model, and the LSTM model is obtained through iterative training.

[0084] (3) Use the trained LSTM model to predict the future compressed feature matrix X pred。

[0085] (4) Using the trained AE model to restore the future compressed feature matrix to the future unstable trend prediction snapshot

[0086] In this embodiment, the transient temperature response of the maximum temperature point is obtained, such as Figure 2 As shown, the temperature fluctuation and temperature range can be accurately obtained; the temperature field calculation results are shown in Figure 3 As shown, the temperature data of the entire power module can be obtained, including the temperature of components such as chips, substrates, and solders.

[0087] In an embodiment of the present invention, an electronic device is also provided, including: one or more processors; a storage device on which one or more programs are stored; when the one or more programs are executed by the one or more processors, the one or more processors implement the power module temperature field hybrid model prediction method of any of the above embodiments.

[0088] In an embodiment of the present invention, a computer-readable storage medium is further provided, on which a computer program is stored. When the program is executed by a processor, the steps in the hybrid model prediction method for the power module temperature field in the above embodiment are implemented.

[0089] The basic principles, main features, and advantages of the present invention are shown and described above. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions are merely illustrative of the principles of the present invention. Various changes and modifications may be made to the present invention without departing from the spirit and scope of the present invention, and such changes and modifications fall within the scope of the invention as claimed.

Claims

1. A power module temperature hybrid model prediction method, characterized in that: The steps include: Step 1: Build a three-dimensional finite element thermal model of the power module and define material properties and boundary conditions; Step 2: Obtain the junction-to-case thermal resistance response curve of the power module through finite element simulation and compare it with the junction-to-case thermal resistance curve in the data sheet to verify the accuracy of the 3D finite element thermal model. Step 3: Obtain sampling snapshots of the power module temperature field through finite element simulation, where the sampling interval is Δt and the number of sampling snapshots is n; Step 4: Based on the dynamic mode decomposition method and deep learning AE-LSTM model, build a hybrid prediction model of the temperature field of the power module; Step 5: Extract the evolution mode of the temperature field by the dynamic mode decomposition method, screen the stable mode and unstable mode according to the frequency and growth rate of the evolution mode, and obtain the stable trend under the stable mode and the unstable trend under the unstable mode; Step 6: Use the recursive method to predict future stable trend snapshots, and use the deep learning AE-LSTM model to predict future unstable trend snapshots; Step 7: Superimpose the future stable trend snapshot and the future unstable trend snapshot to obtain the future temperature field snapshot.

2. The power module temperature hybrid model prediction method according to claim 1, characterized in that: In step 1, the three-dimensional finite element thermal model of the power module is constructed in the following sub-steps: Step 1.1: Based on the internal structural features and geometric dimensions of the power module, use 3D drawing software to create a 3D physical model of the power module in actual proportion. Step 1.2: Mesh the 3D physical model of the power module and set the boundary conditions and material properties of the 3D physical model; Step 1.3: Set the power module chip as the heat source and set the boundary condition to power loss.

3. The power module temperature hybrid model prediction method according to claim 1, characterized in that: In step 4, the hybrid prediction model of the temperature field of the power module is constructed by the following sub-steps: Step 4.1: Obtain a sampling snapshot of the power module temperature field based on finite element simulation and construct the temperature data matrix T as the input of the prediction model: T=[T1 T2...T i ...T n ] Among them, T i Represents the snapshot data of the i-th temperature field, i = 1, 2, ..., n; Step 4.2: Based on the temperature data matrix T, the dynamic mode decomposition method is used to obtain the dynamic mode matrix of the temperature field And the mode coefficient matrix λ: λ=[λ1,λ2,...,λ r ] Among them, λ j 、 denote the jth coefficient and dynamic mode respectively.

4. The power module temperature hybrid model prediction method according to claim 3, characterized in that: In step 5, the frequency f of each temperature field evolution mode is calculated according to the mode coefficient matrix j and growth rate g j , select frequency f j and growth rate g j The mode with all values ​​set to 0 is considered the stable mode. Based on the sampling snapshot and stable mode, the stable trend matrix is ​​obtained by iteration. Based on the temperature data matrix T, obtain the unstable trend matrix in, represents the kth stable mode, λ sj represents the coefficient corresponding to the jth dominant mode, and T1 represents the initial temperature field snapshot.

5. The power module temperature hybrid model prediction method according to claim 4, characterized in that: The frequency f of the temperature field evolution mode j and growth rate g j , calculated by the mode coefficient: Among them, the Re function represents the real part of the complex number, and the Im function represents the imaginary part of the complex number.

6. The power module temperature hybrid model prediction method according to claim 4, characterized in that: In step 6, use the recursive method to predict the future stable trend Using deep learning AE-LSTM model to predict future unstable trends 7. The power module temperature hybrid model prediction method according to claim 6, characterized in that: In step 7, the future stable trend snapshot and the future unstable trend snapshot are superimposed to obtain the future temperature field snapshot T pred :

8. The power module temperature hybrid model prediction method according to claim 5, characterized in that: The deep learning AE-LSTM model is constructed in the following sub-steps: Step 6.1: Unstable trend matrix As the training data of the AE model, the compressed feature matrix X of the unstable trend matrix is ​​obtained using the trained AE model; Step 6.2: Use the compressed feature matrix X as the training data for the LSTM model and obtain the trained LSTM model through iterative training. Step 6.3: Use the trained LSTM model to predict the future compressed feature matrix X pred ; Step 6.4: Use the trained AE model to convert the future compressed feature matrix X pred Revert to a snapshot of the future unstable trend forecast 9. An electronic device, characterized in that: include: one or more processors; a storage device having one or more programs stored thereon; When the one or more programs are executed by the one or more processors, the one or more processors implement the power module temperature hybrid model prediction method according to any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that A computer program is stored thereon, and when the program is executed by a processor, the steps in the power module temperature hybrid model prediction method according to any one of claims 1 to 8 are implemented.

Images