IGBT module transient thermal resistance curve fitting method

Abstract

The invention provides an IGBT (Insulated Gate Bipolar Translator) module transient thermal resistance curve fitting method, which relates to the technical field of semiconductor device testing and comprises the following steps of: actually measuring and acquiring transient thermal resistance data of an IGBT module; performing Kalman filtering smoothing on the transient thermal resistance data to obtain smoothed data; and constructing a fourth-order polynomial function to fit the smoothed data to obtain a transient thermal resistance curve fitting result. The method has the advantages that on the basis of IGBT module transient thermal resistance data obtained through actual measurement, data forward filtering and backward smoothing are carried out by introducing Kalman filtering, external interference noise existing in actual measurement is effectively restrained, and real characteristics of a curve are reserved; by combining nonlinear least square and fourth-order polynomial function fitting, the complex nonlinear dynamic state of the thermal resistance curve is accurately captured, a set of complete data processing and model optimization process is constructed, the smoothness, the accuracy and the robustness of the transient thermal resistance fitting curve are remarkably improved, and the transient thermal resistance fitting method has a wide application prospect. And a more reliable technical support is provided for thermal characteristic analysis, state monitoring and service life evaluation of the IGBT module.

Description

Technical Field

[0001] This invention relates to the field of semiconductor device testing technology, and in particular to a method for fitting transient thermal resistance curves of IGBT modules. Background Technology

[0002] With the rapid development of power semiconductor technology, Insulated Gate Bipolar Transistor (IGBT) modules, due to their high voltage withstand capability, high current capacity, low loss, and fast switching characteristics, have become core power semiconductor devices in medium- and high-power fields such as new energy power generation (e.g., photovoltaic inverters, wind power converters), smart grids (flexible DC transmission, reactive power compensation devices), industrial frequency conversion (motor drive systems), and electric vehicles (vehicle controllers, on-board chargers). As the core of power conversion and control, the long-term operational reliability of IGBT modules is directly related to the safety and efficiency of the entire system. Performance degradation or sudden failures of IGBT modules often lead to system shutdowns, equipment damage, and even safety accidents, causing significant economic losses.

[0003] In real-world operating conditions, IGBT modules generate a significant amount of heat due to power losses, leading to a rapid increase and drastic fluctuation in junction temperature. This can cause chip aging, bond wire detachment, and other malfunctions, which are key factors limiting their lifespan. Therefore, accurate characterization and real-time monitoring of the thermal characteristics of IGBT modules are crucial for improving system reliability.

[0004] The transient thermal resistance curve is a core indicator describing the internal thermal dynamic response characteristics of an IGBT module, reflecting the temperature change process from junction to case under step power. By analyzing this curve, key thermal parameters can be extracted, an equivalent thermal model can be constructed, and it can be used for online junction temperature estimation and lifetime prediction.

[0005] Currently, to obtain accurate transient thermal resistance curves from measured data, mathematical methods are typically used to fit discrete measurement points. Commonly used fitting methods include:

[0006] 1. Exponential Fitting Method: This method is based on the Cauer or Foster thermal network model and uses the superposition of multiple exponential functions to characterize the thermal relaxation process with multiple time constants. Theoretically, it has a high degree of matching with the physical mechanism of heat transfer in IGBT modules, thus resulting in relatively high fitting accuracy. However, the choice of model order is highly dependent on engineering experience and prior knowledge, and it is quite sensitive to measurement noise. Even small noise interference can cause drastic fluctuations in the coefficients of the exponential terms, causing the fitting results to deviate from the true thermal characteristics.

[0007] 2. Piecewise linear fitting method: This method divides the entire transient thermal resistance curve into multiple linear segments according to the time scale, and approximates the thermal response process by fitting a piecewise straight line. It has the advantages of simple algorithm, high computational efficiency, and low engineering implementation difficulty. However, it has shortcomings in terms of curve smoothness and derivative continuity. It is prone to obvious step distortion at the segment nodes, making it difficult to accurately describe the dynamic details of thermal behavior.

[0008] 3. Traditional least squares fitting method: This method determines the fitting parameters by minimizing the sum of squared residuals between the model's predicted values ​​and the measured data, and is a widely used general fitting method. However, in the processing of transient thermal resistance data, it is prone to overall fitting deviation or local overfitting, especially lacking robustness to abnormal fluctuations in the data, and the fitting degree needs to be improved.

[0009] In summary, existing fitting methods generally suffer from some common problems: they are sensitive to noise and outliers in the original data, resulting in highly susceptible and unstable fitting results; they have many model parameters, complex optimization processes, and low computational efficiency; their fitting accuracy is inconsistent across different temperature zones or time scales, and their generalization ability is insufficient; and their processes do not systematically integrate data processing and fitting optimization, making it difficult to achieve high-precision and robust automatic fitting in engineering applications.

[0010] Therefore, we continue to develop a transient thermal resistance curve fitting method that balances data noise reduction capability with nonlinear fitting accuracy. Summary of the Invention

[0011] To address the problems existing in the prior art, this invention provides a method for fitting the transient thermal resistance curve of an IGBT module, comprising:

[0012] Step S1: Obtain the transient thermal resistance data of the IGBT module by actual measurement;

[0013] Step S2: Perform Kalman filtering on the transient thermal resistance data to obtain smoothed data;

[0014] Step S3: Construct a fourth-order polynomial function to fit the smoothed data to obtain the transient thermal resistance curve fitting result of the IGBT module. The expression of the fourth-order polynomial function is as follows:

[0015]

[0016] Where r1~r4 and τ1~τ4 are the positive-zero fitting coefficients obtained by fitting using the nonlinear least squares method, and r1+r2+r3+r4=Z th_jc Z th_jc This is the steady-state value of the thermal resistance.

[0017] Preferably, step S1 includes:

[0018] Step S11: Apply a step power to the IGBT module using a power cycling test bench, while controlling the substrate temperature of the IGBT module using water cooling and measuring the real-time substrate temperature using thermocouples.

[0019] Step S12: Record the real-time junction temperature of the chip corresponding to the built-in chip of the IGBT module from the start of the application of the step power to the process of thermal equilibrium.

[0020] Step S13: Calculate multiple discrete transient thermal resistance data based on the real-time junction temperature of the chip and the real-time temperature of the substrate.

[0021] Preferably, in step S11, the real-time temperature of the substrate is obtained by measuring the calibration curve of the saturation voltage and temperature of the chip built into the IGBT module.

[0022] Preferably, in step S3, the transient thermal resistance data is calculated according to the following formula:

[0023]

[0024] Among them, Z th_jc (t) represents the transient thermal resistance data at time t, where T j (t) represents the real-time junction temperature of the chip at time t, where T is the junction temperature at time t. c (t) represents the real-time temperature of the substrate at time t, P loss The power loss of the applied step power.

[0025] Preferably, before performing step S2, the transient thermal resistance data is further preprocessed, and the preprocessing process includes:

[0026] Step A1: Arrange the transient thermal resistance data in chronological order to form a two-dimensional data table;

[0027] Step A2: Remove abrupt and drifting outliers from the two-dimensional data table to obtain the data table after removal;

[0028] Step A3: Extract the effective range of transient thermal resistance change data from the data table after elimination to obtain the preprocessed data.

[0029] Preferably, the transient thermal resistance data is a sequence of discrete thermal resistance data arranged in chronological order; step S2 includes:

[0030] Step S21: For each transient thermal resistance data, using the observation data at the current time point and before, recursively calculate the state estimate and its covariance at each time point, and then perform forward filtering on the thermal resistance data sequence to obtain the filtered data sequence.

[0031] Step S22: For the filtered data sequence, using the observation information of the entire time series, starting from the last time point, reverse the process to correct and optimize the state estimate at each time point, thereby achieving backward smoothing of the filtered data sequence and finally obtaining the smoothed data.

[0032] Preferably, in step S3, the fitting coefficients obtained by fitting using the nonlinear least squares method include initial parameter estimation and setting parameter boundaries, followed by optimization using the nonlinear least squares method to find the optimal parameters as the fitting parameters.

[0033] Preferably, the optimization of finding the optimal parameters using the nonlinear least squares method includes:

[0034] The initial parameters are iteratively updated by minimizing the sum of squared residuals between the model predictions and the smoothed data. The iteration stops when the number of iterations reaches a preset number of optimization iterations or when the change in the initial parameters is less than a preset parameter change limit and the changed parameter values ​​are within the parameter boundaries, thus obtaining the fitted parameters.

[0035] Preferably, the method further includes generating a process data analysis graph for visualization during the execution of steps S1 to S3, and visualizing the fitting result of the transient thermal resistance curve after the execution of step S3.

[0036] The above technical solution has the following advantages or beneficial effects: Based on the measured transient thermal resistance data of IGBT modules, Kalman filtering is introduced for forward filtering and backward smoothing of the data, which effectively suppresses external interference noise in the measurement and preserves the true characteristics of the curve; combined with nonlinear least squares and fourth-order polynomial function fitting, the complex nonlinear dynamics of the thermal resistance curve are accurately captured, and a complete data processing and model optimization process is constructed, which significantly improves the smoothness, accuracy and robustness of the transient thermal resistance fitting curve, and provides more reliable technical support for the thermal characteristic analysis, condition monitoring and life assessment of IGBT modules. Attached Figure Description

[0037] Figure 1 A flowchart illustrating a method for fitting transient thermal resistance curves of an IGBT module, as a preferred embodiment of the present invention.

[0038] Figure 2 This is a schematic diagram of a sub-process of step S1 in a preferred embodiment of the present invention.

[0039] Figure 3 This is a schematic diagram of the preprocessing process in a preferred embodiment of the present invention;

[0040] Figure 4This is a schematic diagram of a sub-process of step S2 in a preferred embodiment of the present invention;

[0041] Figure 5 In a preferred embodiment of the present invention, a point plot of transient thermal resistance data obtained by actual measurement is shown.

[0042] Figure 6 In a preferred embodiment of the present invention, a comparison chart of transient thermal resistance data smoothing and fitting results is shown.

[0043] Figure 7 In a preferred embodiment of the present invention, a residual analysis diagram of transient thermal resistance data is provided.

[0044] Figure 8 In a preferred embodiment of the present invention, a smoothing error analysis diagram of transient thermal resistance data is provided.

[0045] Figure 9 In a preferred embodiment of the present invention, a decomposition diagram of the transient thermal resistance data fitting curve is shown.

[0046] Figure 10 In a preferred embodiment of the present invention, a transient thermal resistance data distribution diagram is shown. Detailed Implementation

[0047] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. The present invention is not limited to this embodiment; other embodiments that conform to the spirit of the present invention may also fall within the scope of the present invention.

[0048] In a preferred embodiment of the present invention, based on the above-mentioned problems existing in the prior art, a method for fitting the transient thermal resistance curve of an IGBT module is provided, such as... Figure 1 As shown, it includes:

[0049] Step S1: Obtain the transient thermal resistance data of the IGBT module by actual measurement;

[0050] Step S2: Perform Kalman filtering on the transient thermal resistance data to obtain smoothed data;

[0051] Step S3: Construct a fourth-order polynomial function to fit the smoothed data to obtain the transient thermal resistance curve fitting result of the IGBT module. The expression of the fourth-order polynomial function is as follows:

[0052]

[0053] Where r1~r4 and τ1~τ4 are the positive-zero fitting coefficients obtained by fitting using the nonlinear least squares method, and r1+r2+r3+r4=Z th_jc Z th_jc This is the steady-state value of the thermal resistance.

[0054] Specifically, in this embodiment, by performing bidirectional Kalman filtering on the measured transient thermal resistance data, and then adopting a full-link design with fourth-order polynomial optimization fitting, the transient thermal resistance curve is restored with high fidelity, providing accurate data support for IGBT thermal characteristic analysis, junction temperature estimation and lifetime prediction.

[0055] The accuracy of the transient thermal resistance data directly determines the validity of the subsequent fitting results. In this embodiment, a precise temperature control and multi-dimensional monitoring measurement system ensures that the original data closely approximates the true thermal response characteristics of the IGBT, avoiding system errors introduced by external interference. Specifically, for example... Figure 2 As shown, step S1 includes:

[0056] Step S11: Apply a step power to the IGBT module using a power cycling test bench, while controlling the substrate temperature of the IGBT module using water cooling and measuring the real-time substrate temperature using thermocouples.

[0057] Step S12: Record the real-time junction temperature of the chip corresponding to the built-in chip of the IGBT module from the start of the application of the step power to the thermal equilibrium process.

[0058] Step S13: Calculate multiple discrete transient thermal resistance data based on the real-time junction temperature of the chip and the real-time temperature of the substrate.

[0059] Specifically, in this embodiment, the substrate temperature of the IGBT module is controlled by water cooling, which effectively isolates the interference of ambient temperature fluctuations on thermal resistance calculation. Preferably, a thermocouple is placed close to the temperature measurement point on the substrate to achieve real-time acquisition of substrate temperature data, ensuring the capture of subtle dynamic changes in temperature.

[0060] Furthermore, starting from the instant the step power is applied, the chip junction temperature is continuously recorded until thermal equilibrium is reached (typically 10-15 seconds). In a preferred embodiment of the present invention, in step S11, the real-time temperature of the substrate is obtained by measuring the calibration curve of the saturation voltage and temperature of the chip built into the IGBT module. Specifically, the VCE(sat)-Tj calibration curve of this type of IGBT chip is pre-calibrated using a constant temperature chamber. In actual measurement, the chip saturation voltage is synchronously acquired using a high-speed data acquisition card, and the junction temperature is calculated in real time in combination with the calibration curve. At the same time, the substrate temperature measured by a thermocouple can be used as an auxiliary verification to ensure the reliability of the junction temperature data. Compared with single thermocouple monitoring, this scheme improves the response speed to the microsecond level and can accurately capture the instantaneous jump process of junction temperature under step power.

[0061] In a preferred embodiment of the present invention, in step S3, the transient thermal resistance data is calculated according to the following formula:

[0062]

[0063] Among them, Zth_jc (t) represents the transient thermal resistance data at time t, where T j (t) represents the real-time junction temperature of the chip at time t, T c (t) represents the real-time temperature of the substrate at time t, P loss The power loss of the applied step power.

[0064] In a preferred embodiment of the present invention, before performing step S2, the transient thermal resistance data is further preprocessed, such as... Figure 3 As shown, the preprocessing process includes:

[0065] Step A1: Arrange the transient thermal resistance data in chronological order to form a two-dimensional data table;

[0066] Step A2: Remove abrupt and drifting outliers from the two-dimensional data table to obtain the data table after removal;

[0067] Step A3: Extract the effective range of transient thermal resistance change data from the data table after data removal to obtain the preprocessed data.

[0068] Specifically, since measured data is susceptible to electromagnetic interference, sensor contact jitter, etc., which can introduce abrupt or drifting anomalies, in this embodiment, a data preprocessing step is added before performing Kalman filtering. In step A2, a segmented-residual dual threshold strategy is preferred to automatically identify and remove abrupt and drifting anomalies in the two-dimensional data table to avoid abnormal data from polluting subsequent processing.

[0069] Based on the inherent characteristic of IGBT transient thermal resistance rising rapidly and then converging smoothly, the data is divided into two segments with differentiated threshold configurations to address the poor adaptability of a single threshold. Specifically, the execution process of the segmented-residual dual-threshold strategy includes:

[0070] 1. Divide the transient thermal resistance change of the IGBT module into two segments:

[0071] Specifically, t≤1s is defined as the rising segment of IGBT transient thermal resistance, and t>1s is defined as the stable segment of transient thermal resistance.

[0072] 2. For the transient thermal resistance rise phase

[0073] The rated thermal resistance varies greatly during this stage. If a fixed absolute threshold is used, it is easy to miss small anomalies in the low-value range and misjudge normal fluctuations in the high-value range. Therefore, in this embodiment, a relative threshold Z is configured for this stage. threshold

[0074] Z threshold =k*σ residual

[0075] Where k is the residual coefficient, determined by the final steady-state thermal resistance. The larger the steady-state thermal resistance, the higher the peak value of the thermal resistance in the rising segment, and the value of k increases accordingly to ensure that the threshold dynamically adapts to the magnitude of the thermal resistance; σ residual The residual standard deviation is preferably calibrated using the first 100 initial data points without anomalies, reflecting external interference, such as when electromagnetic interference is strong. residual Increase the threshold, and it will automatically widen to reduce false positives.

[0076] As a preferred implementation, if the residual of a single data point is greater than the relative threshold, and the residuals of the preceding 3 data points and the following 3 data points are all less than half of the relative threshold, then the data point can be considered as an instantaneous jump point with no trend continuation, and the data point can be identified as a sudden anomaly.

[0077] In a preferred implementation, if the residuals of three or more consecutive data points are all greater than the relative threshold, the data change is considered to be a trend deviation, and the corresponding data points can be identified as drifting anomalies.

[0078] It is understood that the above embodiments are merely examples and are not intended to limit the present invention.

[0079] 3. For the stable section of transient thermal resistance

[0080] During this stage, the thermal resistance fluctuates slightly around the steady-state value, with small and stable changes. Therefore, in this embodiment, an absolute threshold Z is directly configured during this stage. threshold

[0081] Z threshold =σ residual

[0082] As a preferred implementation, if the residual of a single data point is greater than the absolute threshold, and the residuals of the preceding 3 data points and the following 3 data points are all less than half of the absolute threshold, then the data point can be considered as an instantaneous jump point with no trend continuation, and the data point can be identified as a sudden anomaly.

[0083] As a preferred implementation, if the residuals of three or more consecutive data points are all greater than the absolute threshold, the data change is considered to be a trend deviation, and the corresponding data points can be identified as drifting anomalies.

[0084] 4. Data extraction

[0085] The effective range of transient thermal resistance change data is typically extracted from 0.001s to 10s. Before 0.001s, the power excitation rises, and the data is prone to distortion; after 10s, the IGBT has reached thermal equilibrium, and the thermal resistance shows no significant change. By extracting data between these two points, the extracted data covers the complete thermal response process while eliminating invalid data, effectively improving subsequent processing efficiency.

[0086] To address the residual random noise in the preprocessed data, this embodiment employs a bidirectional Kalman filtering strategy combining forward filtering and backward smoothing. This solves the problem of traditional single-filter noise reduction and trend preservation being difficult to achieve simultaneously, thus realizing a balance between noise suppression and thermal response feature restoration. Specifically, in a preferred embodiment of the invention, the transient thermal resistance data consists of multiple discrete thermal resistance data sequences arranged in chronological order; such as... Figure 4 As shown, step S2 includes:

[0087] Step S21: For each transient thermal resistance data, using the observation data at the current time point and before, recursively calculate the state estimate and its covariance at each time point, and then perform forward filtering on the thermal resistance data sequence to obtain the filtered data sequence.

[0088] Step S22: For the filtered data sequence, using the observation information of the entire time series, the state estimate of each time point is corrected and optimized by recursively working backward from the last time point to achieve backward smoothing of the filtered data sequence, and finally obtain the smoothed data.

[0089] Specifically, in this embodiment, the state equation and observation equation are preferably constructed based on the dynamic characteristics of transient thermal resistance, and the state estimate and covariance at each time point are recursively calculated to achieve preliminary noise filtering. In other words, forward filtering specifically includes: state prediction, uncertainty prediction, calculation of observation difference, calculation of filter gain, state update, and uncertainty update.

[0090] Equations of state:

[0091] x k =f(x) k-1 )+w k =Fx k-1 +w k

[0092]

[0093] Observation equation:

[0094] z k =h(x k )+v k =Hx k +v k

[0095] H = [1 0]

[0096] Where, x k The actual temperature value at time k; z k The measured temperature value at time k; w k For process noise; v kdenoted as observation noise; F is the state transition matrix; H is the observation matrix.

[0097] Given that the overall trend of transient thermal resistance data is that it first rises and then converges to a steady state, a process noise model suitable for transient thermal resistance changes is established, and the observation noise is adaptively estimated based on the data variance.

[0098] Predicting the covariance matrix:

[0099] P k|k-1 =FP k-1|k-1 F T +Q

[0100] Observation covariance matrix:

[0101] S k =HP k|k-1 H T +R

[0102] Where Q is the process noise, which is adaptively adjusted based on the rate of change of the state estimate; and R is the observation noise, which is adaptively adjusted based on the data variance.

[0103] Kalman gain:

[0104]

[0105] Status Update:

[0106]

[0107] Covariance update:

[0108] P k|k =(IK k H)P k|k-1

[0109] Backward smoothing specifically includes calculating the smoothing gain and updating the smoothing state. Building upon forward filtering, the backward smoothing process utilizes transient thermal resistance observations from the entire time series, recursively working backward from the last time point to correct and optimize the state estimate at each time point, thereby obtaining the optimal smoothing estimate across the entire time series.

[0110] Smoothing gain optimization:

[0111] C k =P k|k F T (P k+1|k ) -1

[0112] Smooth state estimation:

[0113]

[0114] The output is a two-dimensional sequence of transient thermal resistance data after smoothing. This sequence preserves the trend and characteristics of the original data and reduces interference from external noise and measurement errors, resulting in a higher signal-to-noise ratio.

[0115] In a preferred embodiment of the present invention, step S3, obtaining a fitting coefficient greater than zero using the nonlinear least squares method includes initial parameter estimation and setting parameter boundaries, followed by optimization using the nonlinear least squares method to find the optimal parameters as fitting parameters.

[0116] In a preferred embodiment of the present invention, the optimization of finding the optimal parameters using the nonlinear least squares method includes:

[0117] The initial parameters are iteratively updated by minimizing the sum of squared residuals between the model's predicted values ​​and the smoothed data. The iteration stops when the number of iterations reaches the preset number of optimization iterations or when the change in the initial parameters is less than the preset parameter change limit and the changed parameter values ​​are within the parameter boundaries, thus obtaining the fitted parameters.

[0118] Specifically, in this embodiment, the expression for the sum of squared residuals between the model's predicted values ​​and the smoothed data is as follows:

[0119]

[0120] in,

[0121] y i These are actual observed values; These are the model's predicted values; This represents the average of the actual observed values.

[0122] In a preferred embodiment of the present invention, the method further includes generating a process data analysis graph for visualization during the execution of steps S1 to S3, and visualizing the fitting result of the transient thermal resistance curve after the execution of step S3.

[0123] Specifically, to visually verify the processing effectiveness of each stage, this method generates multi-dimensional process data analysis diagrams throughout the entire process, such as... Figures 5 to 10 As shown, it specifically includes:

[0124] Comparison of data smoothing and fitting results: By overlaying the original data, smoothed data and fitting curve, the noise suppression effect and curve fit can be clearly shown. It can be intuitively observed that the fitting curve completely follows the trend of thermal resistance change without significant deviation.

[0125] Residual analysis plot: shows the distribution of the fitted residuals, proving that the fitting bias is uniform and small;

[0126] Smoothing error analysis graph: Quantizes the error changes of forward filtering and backward smoothing to verify the advantages of bidirectional filtering;

[0127] Decomposition plot of fitted curve: shows the contribution of each term of the fourth-order polynomial, and clarifies the role of higher-order terms in the smooth convergence of the curve;

[0128] Data distribution plot: Statistically analyze the distribution characteristics of data at each stage to verify the improvement of data quality by preprocessing and filtering.

[0129] In summary, this method optimizes the entire process using measured transient thermal resistance data of IGBT modules. By introducing Kalman filtering for forward and backward smoothing, it effectively suppresses external interference noise in the measurements while preserving the true characteristics of the curves. Combining nonlinear least squares and fourth-order polynomial function fitting, it accurately captures the complex nonlinear dynamics of the thermal resistance curves. A complete data processing and model optimization workflow is constructed, significantly improving the smoothness, accuracy, and robustness of the transient thermal resistance fitting curves. This provides more reliable technical support for the thermal characteristic analysis, condition monitoring, and life assessment of IGBT modules.

[0130] The above description is merely a preferred embodiment of the present invention and does not limit the implementation and protection scope of the present invention. Those skilled in the art should realize that any equivalent substitutions and obvious changes made using the content of this specification and illustrations should be included within the protection scope of the present invention.

Claims

1. A method for fitting the transient thermal resistance curve of an IGBT module, characterized in that, include: Step S1: Obtain the transient thermal resistance data of the IGBT module by actual measurement; Step S2: Perform Kalman filtering on the transient thermal resistance data to obtain smoothed data; Step S3: Construct a fourth-order polynomial function to fit the smoothed data to obtain the transient thermal resistance curve fitting result of the IGBT module. The expression of the fourth-order polynomial function is as follows: Where r1~r4 and τ1~τ4 are the positive-zero fitting coefficients obtained by fitting using the nonlinear least squares method, and r1+r2+r3+r4=Z th_jc Z th_jc This is the steady-state value of the thermal resistance.

2. The IGBT module transient thermal resistance curve fitting method according to claim 1, characterized in that, Step S1 includes: Step S11: Apply a step power to the IGBT module using a power cycling test bench, while controlling the substrate temperature of the IGBT module using water cooling and measuring the real-time substrate temperature using thermocouples. Step S12: Record the real-time junction temperature of the chip corresponding to the built-in chip of the IGBT module from the start of the application of the step power to the process of thermal equilibrium. Step S13: Calculate multiple discrete transient thermal resistance data based on the real-time junction temperature of the chip and the real-time temperature of the substrate.

3. The IGBT module transient thermal resistance curve fitting method according to claim 2, characterized in that, In step S11, the real-time temperature of the substrate is obtained by measuring the calibration curve of the saturation voltage and temperature of the chip built into the IGBT module.

4. The method for fitting the transient thermal resistance curve of an IGBT module according to claim 2, characterized in that, In step S3, the transient thermal resistance data is calculated according to the following formula: Among them, Z th_jc (t) represents the transient thermal resistance data at time t, where T j (t) represents the real-time junction temperature of the chip at time t, where T is the junction temperature at time t. c (t) represents the real-time temperature of the substrate at time t, P loss The power loss of the applied step power.

5. The method for fitting the transient thermal resistance curve of an IGBT module according to claim 1, characterized in that, Before performing step S2, the transient thermal resistance data is preprocessed, and the preprocessing process includes: Step A1: Arrange the transient thermal resistance data in chronological order to form a two-dimensional data table; Step A2: Remove abrupt and drifting outliers from the two-dimensional data table to obtain the data table after removal; Step A3: Extract the effective range of transient thermal resistance change data from the data table after elimination to obtain the preprocessed data.

6. The method for fitting the transient thermal resistance curve of an IGBT module according to claim 1, characterized in that, The transient thermal resistance data consists of multiple discrete thermal resistance data sequences arranged in chronological order; step S2 includes: Step S21: For each transient thermal resistance data, using the observation data at the current time point and before, recursively calculate the state estimate and its covariance at each time point, and then perform forward filtering on the thermal resistance data sequence to obtain the filtered data sequence. Step S22: For the filtered data sequence, using the observation information of the entire time series, starting from the last time point, reverse the process to correct and optimize the state estimate at each time point, thereby achieving backward smoothing of the filtered data sequence and finally obtaining the smoothed data.

7. The method for fitting the transient thermal resistance curve of an IGBT module according to claim 1, characterized in that, In step S3, the fitting coefficients obtained by fitting with the nonlinear least squares method that are greater than zero include initial parameter estimation and setting parameter boundaries, followed by optimization using the nonlinear least squares method to find the optimal parameters as the fitting parameters.

8. The IGBT module transient thermal resistance curve fitting method according to claim 7, characterized in that, The optimization of finding the optimal parameters using the nonlinear least squares method includes: The initial parameters are iteratively updated by minimizing the sum of squared residuals between the model predictions and the smoothed data. The iteration stops when the number of iterations reaches a preset number of optimization iterations or when the change in the initial parameters is less than a preset parameter change limit and the changed parameter values ​​are within the parameter boundaries, thus obtaining the fitted parameters.

9. The method for fitting the transient thermal resistance curve of an IGBT module according to claim 1, characterized in that, It also includes generating process data analysis graphs for visualization during the execution of steps S1 to S3, and visualizing the fitting results of the transient thermal resistance curve after the execution of step S3.

Images