A method and system for constructing an IGBT junction temperature prediction model

Abstract

The application provides a kind of IGBT junction temperature prediction model construction method and system, it is related to power electronic device thermal management technical field, method includes: obtaining the initial data set of the transient thermal response relationship between the IGBT chip and the NTC temperature measuring point in the IGBT power module;Based on the initial data set, the neural network model is trained, and the nonlinear function relationship between time and thermal impedance is learned and characterized;The preset new time series is input into the trained neural network model, and the optimized thermal impedance data set with continuous time and noise suppression is generated;Based on the data set, the lumped parameter thermal network model is parameter fitted, to identify its thermal resistance and heat capacity parameters, complete the construction of junction temperature prediction model.The application uses neural network as a data optimization tool in the parameter identification process of physical model, only uses the temperature signal of NTC temperature measuring point provided by the module, realizes the accurate, convenient online prediction of IGBT junction temperature.

Description

Technical Field

[0001] This invention relates to the field of thermal management technology for power electronic devices, and in particular to a method and system for constructing an IGBT junction temperature prediction model. Background Technology

[0002] As a core switching device in modern power electronic systems, the IGBT (Insulated Gate Bipolar Transistor) directly affects the reliability, efficiency, and lifespan of the module due to its operating junction temperature. Therefore, accurate monitoring and prediction of IGBT junction temperature is of great significance for system condition assessment, fault early warning, and lifespan management.

[0003] Currently, methods for obtaining IGBT junction temperature are mainly divided into two categories: simulation methods and measurement methods. Simulation methods include thermal network modeling, finite element simulation, and thermistor parameter methods; measurement methods include physical contact methods (such as thermocouples) and optical non-contact methods (such as infrared thermal imagers). However, these methods all have significant limitations in practical engineering applications: contact measurements are highly invasive and easily interfere with system operation; non-contact methods are costly and have strict environmental requirements; while simulation methods rely on accurate model parameters, which are difficult to obtain and hard to adapt to changes in actual operating conditions.

[0004] To facilitate temperature monitoring, IGBT power modules typically integrate an NTC (negative temperature coefficient) thermistor to sense the temperature at a specific point within the module. However, due to the complex heat transfer path between the NTC location and the chip junction, and the influence of factors such as installation method, heat dissipation conditions, and multi-chip coupling thermal effects, a significant nonlinear and time-varying thermal delay and coupling relationship exists between the temperature at the NTC measurement point and the chip junction temperature. Current technology lacks a universal and engineered method for accurately and conveniently inferring the IGBT junction temperature from the temperature at the NTC measurement point within the module, making it difficult to achieve low-cost, high-reliability online junction temperature monitoring in practical systems. Summary of the Invention

[0005] This invention provides a method and system for constructing an IGBT junction temperature prediction model to solve the problem that existing technologies have difficulty accurately characterizing the complex thermal coupling effects between the chip and the NTC, as well as between multiple chips, resulting in insufficient accuracy of the final model and inability to accurately reflect the actual dynamic thermal characteristics. This invention enables high-precision, real-time, and convenient online prediction of IGBT junction temperature.

[0006] This invention provides a method for constructing an IGBT junction temperature prediction model, comprising:

[0007] Obtain an initial dataset to characterize the transient thermal response relationship between the IGBT chip and the NTC temperature measurement point inside the IGBT power module.

[0008] Based on the initial dataset, a preset neural network model is trained to learn and characterize the nonlinear functional relationship between time and thermal impedance.

[0009] A set of pre-defined new time series is input into a trained neural network model to generate a set of optimized thermal impedance datasets that are time-continuous and noise-suppressed, wherein the new time series are more densely distributed in time than the initial dataset.

[0010] Based on the optimized thermal impedance dataset, the parameters of the preset lumped parameter thermal network model are fitted to identify its thermal resistance and heat capacity parameters, thereby completing the construction of the junction temperature prediction model.

[0011] According to the method for constructing the IGBT junction temperature prediction model provided by the present invention, the step of obtaining an initial dataset for characterizing the transient thermal response relationship between the IGBT chip and the NTC temperature measurement point inside the IGBT power module specifically includes:

[0012] Establish a three-dimensional finite element thermal model of the IGBT power module and the water-cooling heat dissipation system;

[0013] A step power excitation was applied to the IGBT chip and diode chip in the three-dimensional finite element thermal model, and transient thermal simulation was performed to obtain the temperature correspondence between the junction temperature of the IGBT chip and the temperature of the NTC temperature measuring point over time.

[0014] Based on the temperature correspondence, transient thermal impedance data is calculated and used as the initial dataset.

[0015] According to the construction method of the IGBT junction temperature prediction model provided by the present invention, the transient thermal impedance data includes the self-thermal impedance of the IGBT chip and the diode chip, as well as the mutual thermal impedance between the IGBT chip and the diode chip.

[0016] According to the construction method of the IGBT junction temperature prediction model provided by the present invention, the neural network model includes an input layer, at least one hidden layer and an output layer;

[0017] The training process of the neural network model includes:

[0018] The time data in the initial dataset is used as the input to the neural network model, and the thermal impedance data in the initial dataset is used as the target output. The internal weight parameters of the neural network model are iteratively adjusted through the backpropagation algorithm until the error between the output value of the neural network model and the target output value converges to below a preset threshold.

[0019] According to the method for constructing the IGBT junction temperature prediction model provided by the present invention, the neural network model is a BP neural network, a recurrent neural network, or a convolutional neural network.

[0020] According to the method for constructing an IGBT junction temperature prediction model provided by the present invention, the step of inputting a set of preset new time series into a trained neural network model to generate a set of time-continuous and noise-suppressed optimized thermal impedance dataset specifically includes:

[0021] For each time point in the new time series, a forward propagation calculation is performed in the trained neural network model, and the output of the calculation is used as a data point in the optimized thermal impedance dataset.

[0022] According to the method for constructing an IGBT junction temperature prediction model provided by the present invention, the step of fitting parameters to a preset lumped parameter thermal network model based on the optimized thermal resistance dataset to identify its thermal resistance and thermal capacity parameters, thereby completing the construction of the junction temperature prediction model, specifically includes:

[0023] The mathematical expression of the lumped parameter thermal network model is defined as a function with thermal resistance and heat capacity as undetermined parameters;

[0024] Using the optimized thermal impedance dataset as a fitting benchmark, the undetermined parameters are adjusted by running an optimization algorithm until the fitting error between the thermal impedance curve calculated by the function and the optimized thermal impedance dataset is minimized. The values ​​of the thermal resistance and thermal capacity parameters are then determined, and the lumped parameter thermal network model with determined parameters is obtained as the junction temperature prediction model.

[0025] According to the construction method of the IGBT junction temperature prediction model provided by the present invention, the lumped parameter thermal network model has a multi-order Foster network structure or a multi-order Cauer network structure.

[0026] According to the method for constructing the IGBT junction temperature prediction model provided by the present invention, the algorithm used for parameter fitting is selected from: particle swarm optimization algorithm, least squares method, genetic algorithm or a combination thereof.

[0027] This invention also provides a system for constructing an IGBT junction temperature prediction model, comprising the following modules:

[0028] The initial dataset acquisition module is used to acquire an initial dataset that characterizes the transient thermal response relationship between the IGBT chip and the NTC temperature measurement point inside the IGBT power module.

[0029] The neural network model training module is used to train a preset neural network model based on the initial dataset, so that the model learns and represents the nonlinear functional relationship between time and thermal impedance.

[0030] An optimized thermal impedance dataset generation module is used to input a set of preset new time series into a trained neural network model to generate a set of time-continuous and noise-suppressed optimized thermal impedance datasets, wherein the new time series are more densely distributed in time than the initial dataset.

[0031] The junction temperature prediction model construction module is used to perform parameter fitting on a preset lumped parameter thermal network model based on the optimized thermal impedance dataset, so as to identify its thermal resistance and thermal capacity parameters and complete the construction of the junction temperature prediction model.

[0032] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method for constructing the IGBT junction temperature prediction model as described above.

[0033] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for constructing an IGBT junction temperature prediction model as described above.

[0034] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements a method for constructing an IGBT junction temperature prediction model as described above.

[0035] The present invention provides a method and system for constructing an IGBT junction temperature prediction model. This involves acquiring an initial dataset characterizing the thermal response relationship between the IGBT chip and the NTC temperature measurement points within the module; training a neural network model based on this initial dataset; using the trained neural network model as input, generating a smooth and dense optimized thermal impedance dataset with a more concentrated set of preset time series data; and finally, fitting the parameters of a preset lumped parameter thermal network model based on this optimized dataset to accurately identify its thermal resistance and thermal capacity parameters, thus completing the final junction temperature prediction model construction. By using a neural network as a data optimization tool in the physical model parameter identification process, the invention solves the problem of inaccurate model parameters caused by sparse and noisy simulation data in traditional methods. It achieves accurate and convenient online prediction of IGBT junction temperature using only the temperature signals from the module's built-in NTC temperature measurement points, significantly improving the reliability and availability of power electronic systems. Attached Figure Description

[0036] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0037] Figure 1 This is one of the flowcharts illustrating the construction method of the IGBT junction temperature prediction model provided by this invention.

[0038] Figure 2 This is a schematic diagram of a three-dimensional finite element thermal model of the IGBT power module and water-cooled heat dissipation system provided by the present invention.

[0039] Figure 3 This is a schematic diagram of the process for generating an optimized thermal impedance dataset using a neural network model, as provided by the present invention.

[0040] Figure 4 This invention provides a lumped parameter thermal network model for the coupling of multiple chips in the first IGBT power module.

[0041] Figure 5 The present invention provides the temperature of the first IGBT chip of the first IGBT power module and the temperature at the NTC temperature measurement point location, obtained through finite element simulation.

[0042] Figure 6 It is obtained through calculation provided by this invention. The transient thermal impedance curve.

[0043] Figure 7 This invention provides a method for... The result of neural network training and Foster function fitting of the transient thermal impedance curve.

[0044] Figure 8 The present invention provides a direct approach without performing neural network processing. The result is obtained by fitting the transient thermal impedance curve.

[0045] Figure 9 This is a schematic diagram of the structure of the IGBT junction temperature prediction model construction system provided by the present invention.

[0046] Figure 10 This is a schematic diagram of the structure of the electronic device provided by the present invention.

[0047] Figure label:

[0048] 1: First IGBT power module; 2: Second IGBT power module; 3: Third IGBT power module; 4: Inlet; 5: Outlet; 6: Diode chip; 7: IGBT chip; 8: Upper copper layer; 9: Copper substrate. Detailed Implementation

[0049] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0050] The present invention will now be described in detail with reference to the accompanying drawings. The specific operation methods in the method embodiments can also be applied to the device embodiments or system embodiments. In the description of the present invention, unless otherwise stated, "at least one" includes one or more. "Multiple" refers to two or more. For example, at least one of A, B, and C includes: A existing alone, B existing alone, A and B existing simultaneously, A and C existing simultaneously, B and C existing simultaneously, and A, B, and C existing simultaneously. In the present invention, " / " means "or". For example, A / B can mean A or B. "And / or" in this document is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone.

[0051] The present invention will now be described in detail with reference to specific embodiments.

[0052] In some specific embodiments of the present invention, such as Figure 1 As shown, this solution provides a method for constructing an IGBT junction temperature prediction model, including:

[0053] Step S100: Obtain an initial dataset to characterize the transient thermal response relationship between the IGBT chip and the NTC temperature measurement point inside the IGBT power module.

[0054] Step S200: Based on the initial dataset, train the preset neural network model to learn and characterize the nonlinear functional relationship between time and thermal impedance;

[0055] Step S300: Input a set of preset new time series into the trained neural network model to generate a set of optimized thermal impedance datasets that are time-continuous and noise-suppressed, wherein the new time series are more densely distributed in time than the initial dataset.

[0056] Step S400: Based on the optimized thermal impedance dataset, perform parameter fitting on the preset lumped parameter thermal network model to identify its thermal resistance and thermal capacity parameters, and complete the construction of the junction temperature prediction model.

[0057] It should be noted that existing junction temperature prediction schemes estimate the junction temperature by measuring NTC thermistors. However, a complex heat transfer path exists between the temperature at the NTC measurement point inside the module and the chip junction temperature. This relationship exhibits significant nonlinearity and dynamic characteristics due to the influence of packaging structure, heat dissipation conditions, and multi-chip thermal coupling effects. Traditional methods calibrate thermal network model parameters through experimental testing, a cumbersome and costly process. Furthermore, the calibration results are heavily dependent on specific operating conditions and have poor generalization ability. In addition, experimental data is often sparse and noisy, and directly using it for parameter fitting can lead to a decrease in model accuracy.

[0058] Therefore, this invention obtains initial data that reflects the true heat transfer characteristics, uses the initial data to train a neural network, and then uses the neural network to intelligently smooth and enhance the preset time series data to generate a high-quality dataset. Finally, based on this dataset, the parameters of the lumped parameter thermal network model are identified with high precision, thereby constructing a reliable junction temperature prediction model.

[0059] This invention essentially provides a method for estimating the junction temperature of IGBTs based on the location of NTC temperature measurement points within a module using a neural network algorithm. This method includes thermal simulation based on a finite element model and optimal parameter finding based on a neural network algorithm. The first part obtains the temperature distribution of the device through finite element simulation and calculates the transient thermal impedance curve. The second part generates a smooth and dense dataset by training a BP neural network model to assist in the inversion of thermal impedance parameters.

[0060] The above steps will be explained in detail below through specific embodiments.

[0061] Step S100: Obtain an initial dataset to characterize the transient thermal response relationship between the IGBT chip and the NTC temperature measurement point inside the IGBT power module.

[0062] In some possible embodiments of the present invention, obtaining the initial dataset for characterizing the transient thermal response relationship between the IGBT chip and the NTC temperature measurement point inside the IGBT power module specifically includes:

[0063] A three-dimensional finite element thermal model of the IGBT power module and the water-cooling heat dissipation system is established; the geometric dimensions and material thermophysical properties of each component are defined. The components include the water-cooling plate, the thermal grease layer of the IGBT power module, the copper substrate, the upper copper layer, the DBC layer, the lower copper layer, the chip solder layer, the diode chip, and the IGBT chip layer, etc.

[0064] A step power excitation was applied to the IGBT chip and diode chip in the three-dimensional finite element thermal model, and transient thermal simulation was performed to obtain the temperature correspondence between the junction temperature of the IGBT chip and the temperature of the NTC temperature measuring point over time.

[0065] Based on the temperature correspondence, transient thermal impedance data is calculated and used as the initial dataset.

[0066] Specifically, this embodiment provides an implementation method for obtaining an initial dataset. By employing finite element simulation, it accurately and realistically simulates the physical process of heat conduction inside complex structures, providing a high-fidelity physical basis for the entire modeling process. This ensures that the source of the initial data is reliable and accurate, providing a fundamental guarantee for the accuracy of all subsequent steps.

[0067] The water cooling system is used to dissipate heat from the components in the IGBT power module. Optionally, the water cooling system can use a water-cooled radiator with internal water channels, such as a copper radiator, and the water flow rate can be set to 20L / min.

[0068] In a preferred embodiment, the specific steps for obtaining the initial dataset include:

[0069] Step 1: Modeling

[0070] Reference Figure 2 First, an accurate three-dimensional finite element thermal model of the IGBT power module and its water-cooling system is established in finite element analysis software (such as ANSYS, COMSOL, etc.). This model needs to accurately define the geometric dimensions and material thermophysical properties of each component, such as the water-cooling plate including the inlet 4 and outlet 5, the diode chip 6, the IGBT chip 7, the upper copper layer 8, and the copper substrate 9.

[0071] Step 2, Simulation:

[0072] Apply a known step power excitation (e.g., a constant power of 1W at t=0) to a chip in the model and perform a transient thermal simulation. The simulation will calculate the temperature change over time for all nodes in the model. Record the junction temperature of the IGBT chip to which the power is applied. and the temperature at the NTC temperature measurement point inside the module. The relationship between temperature and time t.

[0073] Step 3: Calculation

[0074] Based on the definition of thermal impedance, transient thermal impedance data are calculated using formula (1). Transient thermal impedance data characterizes transient thermal impedance. Follow t The changing set of data points constitutes the initial dataset mentioned above:

[0075] (1)

[0076] in, This is transient thermal impedance data. P For the applied step power, The temperature at time t is the temperature at the NTC temperature measuring point inside the module. This refers to the junction temperature of the IGBT chip.

[0077] The temperature data in formula (1) above are as follows: , P represents the step function, obtained through finite element simulation.

[0078] Furthermore, in some possible embodiments of the present invention, the transient thermal impedance data includes the self-thermal impedance of the IGBT chip and the diode chip, as well as the mutual thermal impedance between the IGBT chip and the diode chip.

[0079] Specifically, this embodiment provides an implementation method for transient thermal impedance data. In order to construct an accurate model that reflects the complex heat conduction inside a multi-chip module, when performing the above steps, it is necessary to calculate not only the self-thermal impedance of the individual IGBT chip's heating effect on its own temperature (for IGBT chips) but also... i Apply power and measure the chip i (Regarding the temperature rise), it is also necessary to calculate the mutual thermal resistance (to the chip) of the chip's heat generation affecting the temperature of other chips or NTC temperature measuring points. i Apply power and measure the chip j Or the temperature rise at the NTC measuring point.

[0080] It's worth noting that in high-power IGBT power modules, multiple chips typically operate in parallel. The heat generated by one chip not only raises its own temperature but also is conducted through the substrate to adjacent chips, increasing their temperatures—a phenomenon known as thermal coupling. If mutual thermal impedance is ignored, the constructed model will fail to reflect this coupling effect, leading to significant errors in real-world operating conditions where multiple chips generate heat simultaneously. By comprehensively acquiring self-thermal impedance and mutual thermal impedance data, the subsequently constructed model can fully describe the mutual influence between all heat sources within the module, greatly improving the model's realism and accuracy.

[0081] For example, refer to Figure 2 Taking Infineon's FF1400R17IP4 IGBT power module as an example, a precise three-dimensional physical model is established in a finite element analysis software (such as ANSYS, COMSOL, etc.) including the first, second, and third IGBT power modules and a water-cooling system such as a water-cooling radiator. The three IGBT power modules ( Figure 2The first IGBT power module 1, the second IGBT power module 2, and the third IGBT power module 3 are placed on a water-cooled radiator. The radiator has a water inlet on the upper right and a water outlet on the lower right. The IGBT power modules are model FF1400R17IP4. Each IGBT power module has 12 IGBT chips and 6 diode chips that generate power and heat. The temperature at the center of the chips and the temperature at the NTC resistor location are monitored. One IGBT power module constitutes one bridge arm, divided into an upper and a lower transistor. Two IGBT power modules form a complete H-bridge, used as a DC-DC step-down chopper, operating in frequency doubling chopper mode. Based on the H-bridge's operating state, the heat-generating components of the upper bridge arm are the upper IGBT and the anti-parallel diode of the lower transistor; the heat-generating components of the lower bridge arm are the anti-parallel diode of the upper transistor and the lower IGBT. (Reference) Figure 2 A three-dimensional finite element thermal model of the IGBT power module and water-cooling system was constructed. Corresponding material properties were set, followed by mesh generation and condition settings, and the temperature change curve was simulated. The obtained temperature results can be used to calculate the thermal impedance curve data, i.e., time-response thermal impedance. t With thermal impedance matrix Z The corresponding data is used for subsequent data processing.

[0082] A three-dimensional model of the module is established in finite element analysis software. The input power and simulation conditions are determined, the junction temperature of the simulation model is calculated, and the thermal impedance curves between the multiple chips inside the module and the NTC temperature measurement point are calculated. That is, the model needs to accurately define the geometric dimensions and material thermophysical properties of each component, such as IGBT chip 7, diode chip 6, direct bonded copper (DBC) layer, copper substrate 9, thermal grease, and water-cooled plate including inlet 4 and outlet 5 and upper copper layer 8.

[0083] After the model is built, mesh generation is performed and boundary conditions are set. Then, a known step power excitation (e.g., a constant power of 1W applied at t=0) is applied to one or more chips in the model, and transient thermal simulation is performed. The simulation calculates the temperature changes of all nodes in the model over the entire time domain. The junction temperature of the IGBT chip to which power is applied is recorded. and the temperature at the NTC temperature measurement point inside the module. Over time t Corresponding relationship to changing temperatures.

[0084] Finally, based on the definition of thermal impedance, the transient thermal impedance data is calculated using formula (1). This is achieved by applying different chips (e.g., chip...). i Apply power and measure itself (chip) i ) and other chips (chips) j The self-thermal resistance of each chip can be calculated by analyzing the temperature response at the NTC point or the NTC point. and the mutual thermal resistance between the chips .These Follow t The changing set of data points constitutes the initial dataset. Although this dataset is physically accurate, it typically suffers from problems such as sparse data points and numerical computation noise.

[0085] Step S200: Based on the initial dataset, train the preset neural network model to learn and characterize the nonlinear functional relationship between time and thermal impedance;

[0086] In some possible embodiments of the present invention, the neural network model includes an input layer, at least one hidden layer, and an output layer;

[0087] The training process of the neural network model includes:

[0088] The time data in the initial dataset is used as the input to the neural network model, and the thermal impedance data in the initial dataset is used as the target output. The internal weight parameters of the neural network model are iteratively adjusted through the backpropagation algorithm until the error between the output value of the neural network model and the target output value converges to below a preset threshold.

[0089] Specifically, the initial datasets obtained from finite element simulations are typically sparse (sampled only at certain time points to conserve computational resources) and may contain minor noise from numerical calculations. Directly using such data to fit physical model parameters often yields poor results. Neural networks, however, can learn and uncover the hidden, continuous, and smooth underlying patterns behind these sparse, noisy data points.

[0090] This embodiment provides an implementation method for training a neural network model. First, a neural network with a standard topology is constructed, including, for example, an input layer for receiving data at time t, one or more hidden layers for performing nonlinear transformations, and a layer for outputting transient thermal impedance. The output layer. The type of neural network can be chosen according to specific needs; it can be a classic BP neural network, or other types such as convolutional neural networks (CNNs) or recurrent neural networks (RNNs). Taking a BP neural network as an example, creating and configuring a BP neural network includes an input layer, two hidden layers, and an output layer; t As input, with The output is a smooth and dense dataset; after training, the parameters are inverted using the thermal impedance formula.

[0091] The training process of the neural network model specifically includes: processing the time data in the initial dataset obtained in step S100. tAs input to this network, t Transient thermal impedance data corresponding to time The target output (label) is then used as the target output. Then, through the backpropagation algorithm, based on the error between the predicted output and the target output, the parameters such as weights and biases inside the network are iteratively adjusted using optimization strategies such as gradient descent, so that the error (e.g., mean squared error) between the network's predicted output value and the actual target output value continuously decreases until the error converges to a preset sufficiently small threshold.

[0092] Through backpropagation and error convergence, the neural network model can learn and characterize the highly nonlinear functional relationship between time and transient thermal impedance. A trained neural network can then learn and characterize the relationship between time and transient thermal impedance at any given time point. t Accurately predict the corresponding transient thermal impedance .

[0093] Step S300: Input a set of preset new time series into the trained neural network model to generate a set of optimized thermal impedance datasets that are time-continuous and noise-suppressed, wherein the new time series are more densely distributed in time than the initial dataset.

[0094] In some possible embodiments of the present invention, the step of inputting a set of preset new time series into a trained neural network model to generate a set of time-continuous and noise-suppressed optimized thermal impedance dataset specifically includes:

[0095] For each time point in the new time series, a forward propagation calculation is performed in the trained neural network model, and the output of the calculation is used as a data point in the optimized thermal impedance dataset.

[0096] Specifically, this embodiment provides an implementation method for generating an optimized thermal impedance dataset using a neural network model. First, a completely new set of time series is preset. The key characteristic of this set of time series is that its temporal distribution is denser than that of the initial dataset. For example, if the time points in the initial dataset are 0.1s, 0.2s, 0.3s..., then the new time series could be 0.001s, 0.002s, 0.003s... Then, each time point in this dense time series is input one by one into the trained neural network model, and a forward propagation calculation is performed within the model. Each calculation yields an output result, namely the predicted transient thermal impedance value corresponding to that time point. Collecting all these output results constitutes a set of optimized thermal impedance datasets that are temporally continuous, densely packed with data points, and whose noise is suppressed due to the filtering characteristics of the neural network.

[0097] Since a neural network is a smooth nonlinear function and learns the inherent patterns of the data rather than noise during training, its output prediction sequence naturally forms a smooth curve that is continuous in time and has suppressed noise.

[0098] Based on the transient thermal impedance data obtained from simulation, a smooth and dense dataset is generated using a BP neural network algorithm to assist in parameter inversion using formula (4). This method combines the advantages of both data-driven and model-driven approaches. (Reference) Figure 3 The specific implementation steps include: inputting test data, which is time. t and The corresponding data; create and configure a BP neural network, which includes an input layer, two hidden layers, and an output layer; using t as input, and... This is the output; after training, network prediction is performed to obtain... and Define the model function, to To optimize the calculation for the objective, output the parameters of the lumped-parameter thermal network model. - , - .

[0099] In this step, instead of using the neural network directly for the final prediction, it is used to repair and enhance the data. Subsequent fitting of the physical model parameters requires dense, smooth data to obtain a unique and accurate solution. This step processes the sparse, noisy raw data into dense, smooth, high-quality data, providing data support for the final construction of a high-precision physical model.

[0100] Step S400: Based on the optimized thermal impedance dataset, perform parameter fitting on the preset lumped parameter thermal network model to identify its thermal resistance and thermal capacity parameters, and complete the construction of the junction temperature prediction model.

[0101] In some possible embodiments of the present invention, the step of fitting parameters to a preset lumped-parameter thermal network model based on the optimized thermal impedance dataset to identify its thermal resistance and heat capacity parameters, thereby completing the construction of the junction temperature prediction model, specifically includes:

[0102] The mathematical expression of the lumped parameter thermal network model is defined as a function with thermal resistance and heat capacity as undetermined parameters;

[0103] Using the optimized thermal impedance dataset as a fitting benchmark, the undetermined parameters are adjusted by running an optimization algorithm until the fitting error between the thermal impedance curve calculated by the function and the optimized thermal impedance dataset is minimized. The values ​​of the thermal resistance and thermal capacity parameters are then determined, and the lumped parameter thermal network model with determined parameters is obtained as the junction temperature prediction model.

[0104] Specifically, this embodiment provides an implementation method for constructing a junction temperature prediction model. First, a lumped parameter thermal network model is preset. This thermal network model is based on the thermal-circuit analogy principle and uses resistance (thermal resistance) R ) and capacitor (heat capacity) C The combination of these factors describes the physical process of heat conduction.

[0105] There are three ways to dissipate heat in electronic devices: heat conduction, heat radiation, and heat convection. In power electronic devices, the heat generated by the chip is usually carried away by water-cooled or air-cooled heat sinks to prevent heat accumulation and overheating. Heat conduction is the primary heat dissipation method in this process. Heat conduction is analogous to Ohm's law in circuits; thermal parameters and electrical parameters have an analogous relationship. There is limited research on heat conduction from the temperature at the NTC resistor to the chip temperature. This heat flows within the module through the DBC layer and substrate layer of the IGBT. This invention considers the coupling between different chips within the module, referencing... Figure 4 The constructed thermal network model reflects the thermal relationship between the temperature at the NTC temperature measurement point inside the module and the chip temperature. This model can be a multi-order Foster network or Cauer network structure, and its mathematical expression (transfer function or transient impedance expression) can be written as a network with various orders of... R and C The values ​​are functions with undetermined parameters. Then, using the generated optimized thermal impedance dataset as a fitting benchmark, a parameter optimization algorithm is run to identify the undetermined parameters in the above functions. , , , Commonly used optimization algorithms include Particle Swarm Optimization (PSO), least squares, genetic algorithms, and other fitting or classical optimization algorithms. The goal of the algorithm is to continuously adjust the values ​​of R and C to minimize the fitting error (e.g., the root mean square error of all data points) between the thermal impedance curve calculated by these functions and the curve represented by the optimized thermal impedance dataset. When the algorithm converges, it finds the set of values ​​that minimizes the error. R and CWhen the values ​​are reached, parameter identification is complete. This set of precise R and C parameters, together with the preset network topology, constitutes a final junction temperature prediction model that can accurately describe the dynamic thermal characteristics between the IGBT chip and the NTC.

[0106] It is worth noting that the lumped-parameter thermal network model has a clear physical meaning, and its robustness and interpretability far exceed those of a pure neural network black-box model. Therefore, it is more in line with the logic of engineering applications, and its computational cost is extremely small, making it very suitable for embedding into controllers for real-time computation. Through the above settings in this embodiment, the transformation from high-quality data to an accurate physical model is achieved. By fitting the model to a high-quality dataset, the identified... R and C The parameters can accurately reflect the real physical and thermal properties, and a junction temperature prediction model is obtained that is physically meaningful, highly interpretable, and extremely accurate due to the precise parameter identification. At the same time, it has low computational cost and is suitable for online applications.

[0107] Once the above construction process is completed, the resulting junction temperature prediction model (i.e., the thermal network model with determined parameters) can be deployed to the actual power electronic device controller for online junction temperature estimation. During device operation, the controller collects the temperature of the NTC inside the IGBT power module in real time and calculates the real-time power loss of the IGBT chip based on the device's operating current, voltage, and switching frequency. These two real-time changing quantities are used as inputs and substituted into the constructed junction temperature prediction model for rapid calculation, thus outputting the IGBT chip's junction temperature in real time and accurately. The specific process includes:

[0108] Step 1: Finite element simulation model construction:

[0109] In this step, a finite element simulation model of the system needs to be constructed.

[0110] Step 2, thermal simulation of step power consumption:

[0111] In this step, thermal simulation is performed based on the established thermal network model. See also... Figure 4 The heat source of a single IGBT power module in the figure includes 12 IGBT chips and 6 diode chips. chip1-chip12 represent the 1st to 12th IGBT chips respectively, and chip13-chip18 represent the 1st to 6th diode chips respectively. , ,..., This represents the heat dissipation power of 18 chips. , ,..., This corresponds to the thermal resistance between different chips. , , ..., It refers to the temperature of each chip, specifically the chip junction temperature. This refers to the temperature at the NTC temperature measurement point inside the module. A step power loss was applied to each of the 18 chips, and the temperature at time t at the NTC temperature measurement point inside the module was obtained through simulation. and chip temperature , ,..., .

[0112] Step 3, Transient Impedance Curve:

[0113] The coupling impedance and self-thermal impedance curves between the chips were calculated.

[0114] Specifically, according to the formula:

[0115] (2),

[0116] (3),

[0117] in, Indicates the first m The self-thermal resistance of each chip, Indicates the first n The chip and the first m Coupling impedance between individual chips , They represent t The simulation results at time 1 m , n The temperature of each chip express t The temperature at the NTC temperature measurement point inside the module is obtained through constant simulation. Indicates the first m The heat output of each chip.

[0118] From this, the coupling impedance and self-thermal impedance curves between the chips can be calculated, corresponding to... Figure 4 The impedance in, including , ... , The impedance curve, which shows the impedance changing over time, is used for subsequent data processing.

[0119] Step 4: Find the optimal parameters using a neural network algorithm:

[0120] Specifically, based on the thermal impedance data obtained from simulation, a smooth and dense dataset is generated using a BP neural network algorithm. , ),in, , These represent refined time and impedance data, respectively.

[0121] Step 5: Establishing the thermal network model:

[0122] Specifically, according to the formula:

[0123] The expression for an N-layer Foster heat network is:

[0124] (4),

[0125] in, and These represent the thermal resistance and time constant of each layer, respectively, corresponding to the heat capacity parameters of each layer. .

[0126] Will( , Substitute these parameters into formula (4) to perform parameter fitting and output the parameters of the lumped parameter thermal network model. , .

[0127] Step Six: Junction Temperature Prediction

[0128] By inputting the temperature and power loss data of the NTC temperature measurement point inside the module under actual operating conditions, the chip junction temperature can be obtained.

[0129] Specifically, Figure 4 A thermal network model of multiple chip couplings in an IGBT power module, including the thermal impedance matrix. Z for:

[0130] (5),

[0131] The formula for calculating the chip junction temperature is:

[0132] (6),

[0133] In the formula, , This is an 18×1 vector, representing the heat dissipation power and junction temperature of the 18 chips. ; .

[0134] by Figure 4 In hot networks Taking thermal resistance parameters as an example, this includes temperature curves, transient thermal resistance curves, neural network-intensive processing curves, and thermal resistance parameter fitting curves. Figure 5 The temperatures of the chip and NTC temperature measurement points obtained from finite element simulation. Figure 6 For calculation The transient thermal impedance curve. The transient thermal impedance data is used to train the neural network model. Figure 7 To The result of training a neural network and fitting a Foster function to the transient thermal impedance curve. The corresponding fourth-order Foster thermal network impedance parameters are shown in Table 1.

[0135] Table 1

[0136]

[0137] Figure 8 Without performing neural network processing, directly... The result after fitting the transient thermal impedance curve. Figure 8 The results and Figure 7 By comparing the results, it is clear that the neural network processing in this invention can better characterize thermal properties.

[0138] This invention combines finite element simulation and neural network algorithms to construct a junction temperature prediction model based on the temperature at the NTC temperature measurement point inside the module. Taking FF1400R17IP4 as an example, a simulation model of the IGBT power module and water-cooled heat sink was established, simulating the junction temperature of the switching device and extracting the transient impedance curve of the thermal network. Combined with a neural network algorithm, the thermal network model was trained based on existing data to obtain better model parameters, characterizing the heat conduction process between the NTC and the junction temperature of the device. This method only requires measuring the NTC temperature to predict the junction temperature of the device, greatly facilitating junction temperature measurement and monitoring in practical engineering, helping to improve the fault prevention system of equipment, and providing online feedback of the junction temperature inside the module, demonstrating good practicality. It also helps to establish an online reliability assessment system for IGBT power modules, assisting in dynamically predicting weak modules and guiding component replacement and fault handling, thereby improving power supply availability.

[0139] This invention is applicable to IGBT power modules of different specifications and can be used for accelerator pulse power supplies such as two-pole magnet power supplies, four-pole magnet power supplies, and six-pole magnet power supplies.

[0140] In some specific embodiments of the present invention, such as Figure 9 As shown, this solution provides a system for constructing an IGBT junction temperature prediction model, including:

[0141] The initial dataset acquisition module 10 is used to acquire an initial dataset that characterizes the transient thermal response relationship between the IGBT chip and the NTC temperature measurement point inside the IGBT power module.

[0142] The neural network model training module 20 is used to train a preset neural network model based on the initial dataset, so that the model learns and represents the nonlinear functional relationship between time and thermal impedance.

[0143] The optimized thermal impedance dataset generation module 30 is used to input a set of preset new time series into the trained neural network model to generate a set of optimized thermal impedance datasets that are time-continuous and noise-suppressed, wherein the new time series are more densely distributed in time than the initial dataset.

[0144] The junction temperature prediction model construction module 40 is used to perform parameter fitting on a preset lumped parameter thermal network model based on the optimized thermal impedance dataset, so as to identify its thermal resistance and thermal capacity parameters and complete the construction of the junction temperature prediction model.

[0145] The IGBT junction temperature prediction model construction system provided in this embodiment of the invention has a similar implementation principle and beneficial effects to the IGBT junction temperature prediction model construction method shown in the above embodiments. Please refer to the implementation principle and beneficial effects of the IGBT junction temperature prediction model construction method shown in the above embodiments, which will not be repeated here.

[0146] Figure 10 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 10 As shown, the electronic device may include: a processor 1010, a communications interface 1020, a memory 1030, and a communications bus 1040, wherein the processor 1010, the communications interface 1020, and the memory 1030 communicate with each other through the communications bus 1040. The processor 1010 can call logic instructions in the memory 1030 to execute a method for constructing an IGBT junction temperature prediction model. This method includes: acquiring an initial dataset to characterize the transient thermal response relationship between the IGBT chip and the NTC temperature measurement point inside its IGBT power module; training a preset neural network model based on the initial dataset to learn and characterize the nonlinear functional relationship between time and thermal impedance; inputting a set of preset new time series into the trained neural network model to generate a set of time-continuous and noise-suppressed optimized thermal impedance dataset, wherein the new time series is denser in temporal distribution than the initial dataset; and performing parameter fitting on a preset lumped-parameter thermal network model based on the optimized thermal impedance dataset to identify its thermal resistance and thermal capacity parameters, thereby completing the construction of the junction temperature prediction model.

[0147] Furthermore, the logical instructions in the aforementioned memory 1030 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0148] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the construction method of the IGBT junction temperature prediction model provided by the above methods. The method includes: obtaining an initial dataset for characterizing the transient thermal response relationship between the IGBT chip and the NTC temperature measurement point inside the IGBT power module; training a preset neural network model based on the initial dataset to learn and characterize the nonlinear functional relationship between time and thermal impedance; inputting a set of preset new time series into the trained neural network model to generate a set of time-continuous and noise-suppressed optimized thermal impedance dataset, wherein the new time series is denser in time distribution than the initial dataset; and performing parameter fitting on a preset lumped parameter thermal network model based on the optimized thermal impedance dataset to identify its thermal resistance and thermal capacity parameters, thereby completing the construction of the junction temperature prediction model.

[0149] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program implements a method for constructing an IGBT junction temperature prediction model provided by the above methods. The method includes: acquiring an initial dataset for characterizing the transient thermal response relationship between the IGBT chip and the NTC temperature measurement point inside the IGBT power module; training a preset neural network model based on the initial dataset to learn and characterize the nonlinear functional relationship between time and thermal impedance; inputting a set of preset new time series into the trained neural network model to generate a set of time-continuous and noise-suppressed optimized thermal impedance dataset, wherein the new time series is denser in temporal distribution than the initial dataset; and performing parameter fitting on a preset lumped parameter thermal network model based on the optimized thermal impedance dataset to identify its thermal resistance and thermal capacity parameters, thereby completing the construction of the junction temperature prediction model.

[0150] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0151] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0152] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for constructing an IGBT junction temperature prediction model, characterized in that, include: Obtain an initial dataset to characterize the transient thermal response relationship between the IGBT chip and the NTC temperature measurement points inside its IGBT power module, specifically including: A three-dimensional finite element thermal model of the IGBT power module and the water-cooling heat dissipation system is established; a step power excitation is applied to the IGBT chip and diode chip in the three-dimensional finite element thermal model and transient thermal simulation is performed to obtain the temperature correspondence between the junction temperature of the IGBT chip and the temperature of the NTC temperature measurement point over time; based on the temperature correspondence, transient thermal impedance data is calculated and used as the initial dataset; Based on the initial dataset, a preset neural network model is trained to learn and characterize the nonlinear functional relationship between time and thermal impedance. A set of preset new time series is input into the trained neural network model, wherein the new time series is denser in time distribution than the initial dataset; for each time point in the new time series, a forward propagation calculation is performed in the trained neural network model, and the output of the calculation is used as a data point in the optimized thermal impedance dataset to generate a set of time-continuous and noise-suppressed optimized thermal impedance dataset. Based on the optimized thermal impedance dataset, parameter fitting is performed on the preset lumped parameter thermal network model to identify its thermal resistance and heat capacity parameters, thereby completing the construction of the junction temperature prediction model, specifically including: The mathematical expression of the lumped parameter thermal network model is defined as a function with thermal resistance and heat capacity as undetermined parameters. Using the optimized thermal impedance dataset as a fitting benchmark, the undetermined parameters are adjusted by running an optimization algorithm until the fitting error between the thermal impedance curve calculated by the function and the optimized thermal impedance dataset is minimized. The values ​​of the thermal resistance and heat capacity parameters are then determined, and the lumped parameter thermal network model with determined parameters is obtained, which serves as the junction temperature prediction model.

2. The method for constructing the IGBT junction temperature prediction model according to claim 1, characterized in that, The transient thermal impedance data includes the self-thermal impedance of the IGBT chip and the diode chip, as well as the mutual thermal impedance between the IGBT chip and the diode chip.

3. The method for constructing the IGBT junction temperature prediction model according to claim 1, characterized in that, The neural network model includes an input layer, at least one hidden layer, and an output layer; The training process of the neural network model includes: The time data in the initial dataset is used as the input to the neural network model, and the thermal impedance data in the initial dataset is used as the target output. The internal weight parameters of the neural network model are iteratively adjusted through the backpropagation algorithm until the error between the output value of the neural network model and the target output value converges to below a preset threshold.

4. The method for constructing the IGBT junction temperature prediction model according to claim 1 or 3, characterized in that, The neural network model is a backpropagation neural network, a recurrent neural network, or a convolutional neural network.

5. The method for constructing the IGBT junction temperature prediction model according to claim 1, characterized in that, The lumped parameter thermal network model has a multi-order Foster network structure or a multi-order Cauer network structure.

6. The method for constructing the IGBT junction temperature prediction model according to claim 1, characterized in that, The algorithm used for parameter fitting is selected from: particle swarm optimization algorithm, least squares method, genetic algorithm or a combination thereof.

7. A system for constructing an IGBT junction temperature prediction model, characterized in that, include: The initial dataset acquisition module is used to acquire an initial dataset characterizing the transient thermal response relationship between the IGBT chip and the NTC temperature measurement points inside its IGBT power module. Specifically, it includes: A three-dimensional finite element thermal model of the IGBT power module and the water-cooling heat dissipation system is established; a step power excitation is applied to the IGBT chip and diode chip in the three-dimensional finite element thermal model and transient thermal simulation is performed to obtain the temperature correspondence between the junction temperature of the IGBT chip and the temperature of the NTC temperature measurement point over time; based on the temperature correspondence, transient thermal impedance data is calculated and used as the initial dataset; The neural network model training module is used to train a preset neural network model based on the initial dataset, so that the model learns and represents the nonlinear functional relationship between time and thermal impedance. An optimized thermal impedance dataset generation module is used to input a set of preset new time series into a trained neural network model, wherein the new time series is denser in temporal distribution than the initial dataset; for each time point in the new time series, a forward propagation calculation is performed in the trained neural network model, and the output of the calculation is used as a data point in the optimized thermal impedance dataset to generate a set of time-continuous and noise-suppressed optimized thermal impedance datasets. The junction temperature prediction model construction module is used to perform parameter fitting on a preset lumped parameter thermal network model based on the optimized thermal impedance dataset, in order to identify its thermal resistance and heat capacity parameters, and complete the construction of the junction temperature prediction model. Specifically, it includes: The mathematical expression of the lumped parameter thermal network model is defined as a function with thermal resistance and heat capacity as undetermined parameters. Using the optimized thermal impedance dataset as a fitting benchmark, the undetermined parameters are adjusted by running an optimization algorithm until the fitting error between the thermal impedance curve calculated by the function and the optimized thermal impedance dataset is minimized. The values ​​of the thermal resistance and heat capacity parameters are then determined, and the lumped parameter thermal network model with determined parameters is obtained, which serves as the junction temperature prediction model.

Images