IGBT module failure prediction method and system

Abstract

The invention belongs to the technical field of new energy wind power generation, and discloses an IGBT module failure prediction method and system, and the method comprises the steps: obtaining electric stress data, thermal stress data and mechanical stress data of a to-be-detected IGBT module, carrying out the time-space alignment processing, and obtaining a multi-dimensional feature vector of the to-be-detected IGBT module; performing multi-physical field coupling effect quantitative modeling based on the multi-dimensional feature vector to obtain a coupling feature matrix; the coupling characteristic matrix comprises a corrected electrical-thermal coupling coefficient, a thermal-mechanical coupling coefficient, a normalized electrical-mechanical coupling coefficient, a maximum temperature gradient and strain dominated by mechanical stress; inputting the coupling characteristic matrix into a pre-constructed coupling sensitive failure prediction model, and outputting to obtain a predicted failure cycle index of the to-be-tested IGBT module; according to the method, the accuracy, comprehensiveness and timeliness of IGBT module failure prediction are effectively improved, and a scientific basis is provided for IGBT model selection, life prediction and inverter system optimization.

Description

Technical Field

[0001] This invention belongs to the field of new energy wind power generation technology, and relates to the field of reliability evaluation of IGBT modules in converters, and particularly to an IGBT module failure prediction method and system. Background Technology

[0002] In the field of new energy wind power generation, with the increasing demands for the reliability of power generation systems, the converter, as a key component, is crucial, and the performance of its internal IGBT (Insulated Gate Bipolar Transistor) module is of paramount importance. IGBT modules need to operate stably under various extreme climatic conditions, including the high temperatures of desert summers and the frigid conditions of winter, while also adapting to diverse drive parameters. Extreme environments place extremely high demands on the failure prediction of IGBT modules, as any failure could lead to the shutdown of the entire wind power generation system, resulting in significant economic losses. Therefore, accurately predicting IGBT module failure is of great significance for ensuring the stable operation and efficient power generation of wind power generation systems.

[0003] In practical applications, existing failure prediction methods for IGBT modules typically study electrical stress, thermal stress, and mechanical stress separately, attempting to reveal the failure mechanism from different perspectives. For example, electrical stress (such as the high rate of change of current during switching) can lead to local hot spots; thermal stress (such as junction temperature fluctuations, i.e., the amount of change in junction temperature) can induce solder layer fatigue; and mechanical stress (such as thermal expansion coefficient mismatch) can exacerbate delamination failure between the chip and the substrate. However, existing methods often neglect the interaction between multiple physical fields, and the coupling effect of electrical stress, thermal stress, and mechanical stress is particularly significant under extreme conditions, such as high-frequency switching and high-temperature environments. However, existing technologies lack systematic modeling and quantitative analysis of this effect, resulting in a large deviation between the predicted results and the actual failure situation. The accuracy of their failure prediction cannot meet practical needs. Summary of the Invention

[0004] To address the technical problems existing in the prior art, this invention provides an IGBT module failure prediction method and system, which solves the technical problem that the existing technology lacks systematic modeling and quantitative analysis of the relationship between multiple physical fields, resulting in a large deviation between the prediction results and the actual failure situation, and the accuracy of its failure prediction cannot meet the actual needs.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows: This invention provides a method for predicting IGBT module failures, comprising: The electrical stress data, thermal stress data, and mechanical stress data of the IGBT module under test are acquired and spatiotemporally aligned to obtain the multidimensional feature vector of the IGBT module under test. Based on the multidimensional feature vector of the IGBT module under test, a quantitative model of multiphysics coupling effect is performed to obtain the coupling feature matrix. The coupling feature matrix includes the corrected electro-thermal coupling coefficient, the thermo-mechanical coupling coefficient, the normalized electro-mechanical coupling coefficient, the maximum temperature gradient, and the strain dominated by mechanical stress. The coupling feature matrix is ​​input into a pre-built coupling-sensitive failure prediction model, and the predicted failure cycle number of the IGBT module under test is output. The pre-built coupling-sensitive failure prediction model is a random forest regression model.

[0006] Furthermore, it also includes: Obtain the operating ambient temperature and humidity of the IGBT module under test; The coupling feature matrix, the operating temperature and humidity of the IGBT module under test are input into the pre-built dynamic failure model, and the predicted failure probability of the IGBT module under test is output; wherein, the pre-built dynamic failure model is a trained deep neural network model. The predicted failure probability of the IGBT module under test is compared with the preset warning threshold, and the failure warning signal of the IGBT module under test is output.

[0007] Furthermore, the electrical stress data of the IGBT module under test includes switching transient waveform data, switching loss characteristics, and current change rate characteristics; The thermal stress data of the IGBT module under test includes junction temperature distribution data, heat sink temperature, and maximum temperature gradient; The mechanical stress data of the IGBT module under test includes microstrain data, principal vibration frequency, and strain dominated by mechanical stress.

[0008] Furthermore, the modeling process for the modified electro-thermal coupling coefficient is as follows:

[0009]

[0010] in, This is the corrected electro-thermal coupling coefficient; The defined electro-thermal coupling coefficient; The predetermined ambient temperature; The maximum temperature gradient; For activation losses; To shut down losses; This represents the average value of the test current.

[0011] Furthermore, the modeling process for the thermo-mechanical coupling coefficient is as follows:

[0012]

[0013] in, Thermo-mechanical coupling coefficient; Strain dominated by mechanical stress; The dominant vibration frequency; Characterized by thermal stress; For reference frequency; This represents the maximum value of the temperature time series matrix; This is the temperature timing data for the radiator.

[0014] Furthermore, the modeling process for the normalized electromechanical coupling coefficient is as follows:

[0015]

[0016] in, The normalized electromechanical coupling coefficient; The defined electromechanical coupling coefficient; The current rise rate; This is the short-circuit electromotive force.

[0017] The further calculation process for the maximum temperature gradient is as follows:

[0018] in, The maximum temperature gradient; This is the temperature time series matrix; Corresponding to ( In two-dimensional coordinates The temperature value at time t, where k is the resolution of the spatial temperature distribution field.

[0019] Furthermore, the calculation process for strain dominated by mechanical stress is as follows:

[0020] in, Strain dominated by mechanical stress; This is a time series of strain caused by thermal expansion; The coefficient of thermal expansion of the substrate material. For reference temperature; This is the temperature time series of the radiator.

[0021] The present invention also provides an IGBT module failure prediction system, comprising: The data acquisition and feature extraction module is used to acquire the electrical stress data, thermal stress data and mechanical stress data of the IGBT module under test, and perform spatiotemporal alignment processing to obtain the multidimensional feature vector of the IGBT module under test. The coupling modeling module is used to quantitatively model the multi-physics coupling effect based on the multi-dimensional feature vector of the IGBT module under test, and obtain the coupling feature matrix. The coupling feature matrix includes the corrected electro-thermal coupling coefficient, the thermo-mechanical coupling coefficient, the normalized electro-mechanical coupling coefficient, the maximum temperature gradient, and the strain dominated by mechanical stress. The coupling-sensitive failure prediction module is used to input the coupling feature matrix into the pre-built coupling-sensitive failure prediction model and output the predicted failure cycle number of the IGBT module under test; wherein, the pre-built coupling-sensitive failure prediction model is a random forest regression model.

[0022] Furthermore, it also includes: The environmental data acquisition module is used to acquire the operating temperature and humidity of the IGBT module under test. The failure probability prediction module is used to input the coupling feature matrix, the operating temperature and humidity of the IGBT module under test into a pre-built dynamic failure model, and output the predicted failure probability of the IGBT module under test; wherein, the pre-built dynamic failure model is a trained deep neural network model. The failure warning module is used to compare the predicted failure probability of the IGBT module under test with the preset warning threshold and output a failure warning signal for the IGBT module under test.

[0023] Compared with the prior art, the beneficial effects of the present invention are as follows: The IGBT module failure prediction method provided by this invention quantifies and models multiphysics coupling effects and utilizes a pre-constructed coupling-sensitive failure prediction model for failure prediction. This effectively improves the accuracy, comprehensiveness, and timeliness of IGBT module failure prediction, providing a scientific basis for IGBT module selection, life prediction, and inverter system optimization. It is of great significance for ensuring the stable operation and efficient power generation of wind power systems. Specifically, by acquiring the electrical stress, thermal stress, and mechanical stress data of the IGBT module under test and performing spatiotemporal alignment processing to obtain multidimensional feature vectors, multiphysics coupling effects are quantified and modeled based on these feature vectors. This yields a coupling feature matrix containing the corrected electro-thermal coupling coefficient, thermal-mechanical coupling coefficient, normalized electro-mechanical coupling coefficient, maximum temperature gradient, and strain dominated by mechanical stress. Through systematic modeling and quantitative analysis of multiphysics coupling effects, a comprehensive analysis is conducted. This approach fully considers the interactions between multiple physical fields (electric, thermal, and mechanical), significantly improving the accuracy of IGBT module failure prediction and enabling it to more accurately reflect IGBT module failures during actual operation. Secondly, it utilizes a pre-built coupling-sensitive failure prediction model based on a random forest regression model. The random forest regression model possesses powerful data processing and generalization capabilities, handling complex nonlinear relationships. By inputting the coupling feature matrix containing multi-physics coupling information into the pre-built coupling-sensitive failure prediction model, the advantages of the random forest regression model can be fully utilized. Combined with the impact of multi-physics coupling effects on IGBT module failure, the failure cycle count of the IGBT module under specific operating conditions can be predicted more accurately. This provides a scientific basis for IGBT module selection, lifespan prediction, and inverter system optimization, thereby providing a more reliable basis for the maintenance and operation management of wind power generation systems.

[0024] Furthermore, a pre-trained deep neural network model is used as a pre-built dynamic failure model. This model can automatically learn complex patterns and features in the data. By acquiring the operating temperature and humidity of the IGBT module under test and inputting the coupling feature matrix, operating temperature, and operating humidity into the pre-built dynamic failure model, the influence of multi-physics coupling effects and operating environment factors on IGBT module failure can be comprehensively considered. This allows for a more comprehensive assessment of the failure probability of the IGBT module in actual operation, making the prediction results closer to reality. Secondly, by setting a warning threshold, a warning signal can be issued in a timely manner when the IGBT module is close to failure, reminding maintenance personnel to take appropriate measures to avoid the shutdown of the entire wind power generation system due to IGBT module failure, thereby reducing huge economic losses and ensuring the stable operation and efficient power generation of the wind power generation system.

[0025] The IGBT module failure prediction system provided by this invention has all the advantages of the above-mentioned IGBT module failure prediction methods. Attached Figure Description

[0026] To more clearly illustrate the technical solutions in the embodiments of this application 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 only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0027] Figure 1 A flowchart of the IGBT module failure prediction method provided in Example 1; Figure 2 The flowchart is for the IGBT module failure prediction method provided in Example 2; Figure 3 This is a structural block diagram of the IGBT module failure prediction system provided in Example 3; Figure 4 The diagram shows the structural block diagram of the IGBT module failure prediction system provided in Example 4. Detailed Implementation

[0028] To make the technical problems, technical solutions, and beneficial effects solved by this application clearer, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0029] This invention provides a method for predicting IGBT module failures, comprising the following steps: Step 100: Obtain the electrical stress data, thermal stress data, and mechanical stress data of the IGBT module under test, and perform spatiotemporal alignment processing to obtain the multidimensional feature vector of the IGBT module under test.

[0030] Step 200: Quantitatively model the multi-physics coupling effect based on the multi-dimensional feature vector of the IGBT module under test to obtain the coupling feature matrix; wherein, the coupling feature matrix includes the corrected electro-thermal coupling coefficient, the thermo-mechanical coupling coefficient, the normalized electro-mechanical coupling coefficient, the maximum temperature gradient, and the strain dominated by mechanical stress.

[0031] Step 300: Input the coupling feature matrix into the pre-built coupling-sensitive failure prediction model and output the predicted failure cycle number of the IGBT module under test; wherein, the pre-built coupling-sensitive failure prediction model is a random forest regression model.

[0032] Optionally, the IGBT module failure prediction method further includes: Step 400: Obtain the operating temperature and humidity of the IGBT module under test.

[0033] Step 500: Input the coupling feature matrix, the operating temperature and humidity of the IGBT module under test into the pre-built dynamic failure model, and output the predicted failure probability of the IGBT module under test; wherein, the pre-built dynamic failure model is a trained deep neural network model.

[0034] Step 600: Compare the predicted failure probability of the IGBT module under test with the preset warning threshold, and output the failure warning signal of the IGBT module under test.

[0035] In the above embodiments, by acquiring electrical, thermal, and mechanical stress data and aligning them spatiotemporally to form a multidimensional feature vector, and then quantitatively modeling the multi-physics coupling effect to obtain a coupling feature matrix, the interaction of electrical, thermal, and mechanical multi-physics fields is fully considered, greatly improving the accuracy of failure prediction. Among them, the use of a random forest regression model to process the coupling feature matrix can accurately predict the number of failure cycles, providing a scientific basis for IGBT module selection, life prediction, and inverter system optimization. In addition, combined with the operating environment temperature and humidity, a deep neural network model is used to comprehensively evaluate the failure probability. By comparing the predicted failure probability with the warning threshold and outputting a warning signal, timely reminders to operation and maintenance can be given, avoiding system downtime and huge economic losses caused by IGBT module failure, and effectively ensuring the stable and efficient operation of the wind power generation system.

[0036] The following specific embodiments further explain the IGBT module failure prediction method provided by the present invention: Example 1 As attached Figure 1 As shown, this embodiment 1 provides a method for predicting IGBT module failure, including the following steps: Step 1: Obtain the electrical stress data, thermal stress data, and mechanical stress data of the IGBT module under test, and perform spatiotemporal alignment processing to obtain the multidimensional feature vector of the IGBT module under test.

[0037] Specifically, the process is as follows: Step 11: Obtain the electrical stress data of the IGBT module under test; the electrical stress data of the IGBT module under test includes switching transient waveform data, switching loss characteristics, and current change rate characteristics; specifically, the process is as follows: A dual-pulse testing method was used, employing a high-voltage differential probe to acquire the collector-emitter voltage (V) of the IGBT module under test. CE ) and collector current (I C The waveform is analyzed to obtain a data stream sequence containing the collector-emitter voltage sequence and the collector current sequence corresponding to the acquisition time, which is used as the switching transient waveform data; wherein, the bandwidth of the high-voltage differential probe is greater than or equal to 200MHz, and the sampling frequency is 10MHz; the switching transient waveform data is represented as follows:

[0038] in, For the collected time series, This is the collector-emitter voltage sequence corresponding to the acquisition time. This is the collector current sequence corresponding to the acquisition time.

[0039] Collector-emitter voltage sequence corresponding to the acquisition time in the switching transient waveform data collector current sequence corresponding to the acquisition time Integral calculations are performed to obtain the switching loss characteristics; these characteristics include the turn-on loss E. on and shutdown loss E off All units are in mJ; specifically, the turn-on loss E on and shutdown loss E off The calculation process is as follows:

[0040]

[0041] in, For the conduction time, The single conduction time, For the shutdown time, This refers to the single shutdown time.

[0042] Collector current sequence corresponding to the acquisition time in the switching transient waveform data Perform first-order differential operations to calculate the current rise rate di / dt, which is used as the current change rate characteristic; where the unit of the current rise rate di / dt is A / ns.

[0043] Step 12: Obtain the thermal stress data of the IGBT module under test; the thermal stress data of the IGBT module under test includes junction temperature distribution data, heat sink temperature, and maximum temperature gradient; specifically, the process is as follows: The surface temperature field of the IGBT module under test was captured using an infrared thermal imager, and a temperature time series matrix was generated. This serves as the junction temperature distribution data; among which, the temperature time series matrix... middle Corresponding to ( In two-dimensional coordinates The temperature value at that time, k is the resolution of the spatial temperature distribution field; preferably, the resolution of the spatial temperature distribution field obtained by the infrared thermal imager is 256×256 pixels, that is, k=256.

[0044] An NTC sensor is pre-embedded in the heatsink of the IGBT module under test. Temperature is collected using the embedded NTC sensor to obtain time-series data. [ ] represents the radiator temperature; where the sampling rate of the NTC sensor is 1kHz.

[0045] For temperature time series matrix Spatial gradient calculations are performed to extract the maximum temperature gradient; the calculation process for the maximum temperature gradient is as follows:

[0046] in, The maximum temperature gradient; To find the maximum function; This is the temperature time series matrix; Corresponding to ( In two-dimensional coordinates The temperature value at time t, where k is the resolution of the spatial temperature distribution field.

[0047] Step 13: Obtain the mechanical stress data of the IGBT module under test; the mechanical stress data of the IGBT module under test includes micro-strain data, principal vibration frequency, and strain dominated by mechanical stress; specifically, the process is as follows: A fiber Bragg grating sensor is mounted on the substrate surface of the IGBT module under test, and the strain time series caused by thermal expansion is measured using the fiber Bragg grating sensor. [ ], as micro-strain data; among which, the fiber optic grating sensor has an accuracy of ±1με and a sampling rate of not less than 10kHz.

[0048] strain time series [ Perform a Fast Fourier Transform (FFT) to extract the principal vibration frequency. Among them, the principal vibration frequency The unit is Hz.

[0049] Based on time series data Subtracting the strain component caused by pure thermal expansion, we obtain the strain dominated by mechanical stress. The calculation process for the strain dominated by mechanical stress is as follows:

[0050] in, Strain dominated by mechanical stress; This is a time series of strain caused by thermal expansion; The coefficient of thermal expansion of the substrate material. For reference temperature; This is the temperature time series of the radiator.

[0051] Step 14: Based on GPS time synchronization, inject a unified timestamp signal into the electrical stress data, thermal stress data, and mechanical stress data of the IGBT module under test; and input the temperature time series matrix of the infrared thermal imager. Coordinate registration was performed with the substrate strain sensor position to establish a spatial mapping relationship. Based on the injected unified timestamp signal and the established spatial mapping relationship, the electrical stress data, thermal stress data, and mechanical stress data of the IGBT module under test were aligned to obtain the multidimensional feature vector of the IGBT module under test. The multidimensional feature vector of the IGBT module under test is as follows:

[0052] in, This represents the multidimensional feature vector of the IGBT module under test.

[0053] Step 2: Quantitatively model the multi-physics coupling effect based on the multi-dimensional feature vector of the IGBT module under test to obtain the coupling feature matrix; wherein, the coupling feature matrix includes the corrected electro-thermal coupling coefficient, the thermo-mechanical coupling coefficient, the normalized electro-mechanical coupling coefficient, the maximum temperature gradient, and the strain dominated by mechanical stress.

[0054] Specifically, the steps are as follows: Step 21: Based on the multi-dimensional feature vector of the IGBT module under test, calculate the electro-thermal coupling coefficient to obtain the corrected electro-thermal coupling coefficient; specifically, first define the electro-thermal coupling coefficient; the defined electro-thermal coupling coefficient is as follows:

[0055] in, The defined electro-thermal coupling coefficient, This represents the average value of the test current.

[0056] Secondly, the defined electro-thermal coupling coefficient is dynamically corrected to obtain the corrected electro-thermal coupling coefficient; the dynamic correction process of the corrected electro-thermal coupling coefficient is as follows:

[0057] in, This is the corrected electro-thermal coupling coefficient; This is the electro-thermal coupling coefficient before correction, i.e., the defined electro-thermal coupling coefficient; The predetermined ambient temperature.

[0058] Step 22: Based on the multidimensional feature vector of the IGBT module under test, calculate the thermo-mechanical coupling coefficient to obtain the thermo-mechanical coupling coefficient; the modeling process of the thermo-mechanical coupling coefficient is as follows:

[0059]

[0060] in, Thermo-mechanical coupling coefficient; Strain dominated by mechanical stress; The dominant vibration frequency; Characterized by thermal stress; The reference frequency is set to 1kHz. This represents the maximum value of the temperature time series matrix; This is the temperature timing data for the radiator.

[0061] It should be noted that in the modeling of the thermo-mechanical coupling coefficient, if the strain is dominated by mechanical stress... Greater than 500 με or thermal stress characteristics For temperatures below 10℃, strains dominated by mechanical stress are excluded. and thermal stress characteristics .

[0062] Step 23: Based on the multi-dimensional feature vector of the IGBT module under test, calculate the electro-mechanical coupling coefficient to obtain the normalized electro-mechanical coupling coefficient; specifically, first define the electro-mechanical coupling coefficient; the defined electro-mechanical coupling coefficient is as follows:

[0063] in, The defined electromechanical coupling coefficient; The current rise rate; This is the short-circuit electromotive force.

[0064] Secondly, the defined electro-mechanical coupling coefficient is normalized to restrict it to the [0,1] interval, thus obtaining the normalized electro-mechanical coupling coefficient. The calculation process of the normalized electro-mechanical coupling coefficient is as follows:

[0065] in, This is the normalized electromechanical coupling coefficient.

[0066] Step 24: Fuse the corrected electro-thermal coupling coefficient, thermo-mechanical coupling coefficient, normalized electro-mechanical coupling coefficient, maximum temperature gradient, and strain dominated by mechanical stress to obtain the coupling characteristic matrix; wherein, the coupling characteristic matrix is ​​as follows:

[0067] in, This is the coupling characteristic matrix.

[0068] Step 3: Input the coupling feature matrix into the pre-constructed coupling-sensitive failure prediction model, and output the predicted failure cycle number N of the IGBT module under test. pred Among them, the pre-constructed coupling-sensitive failure prediction model is a random forest regression model.

[0069] It should be noted that the construction process of the coupling-sensitive failure prediction model includes: using the average number of failure cycles N of the IGBT module under test during factory testing. fail Using the target variable, a random forest regression model is trained to obtain a pre-built coupling-sensitive failure prediction model.

[0070] The IGBT module failure prediction method described in Example 1 obtains multi-dimensional feature vectors by comprehensively acquiring electrical stress, thermal stress, and mechanical stress data and performing spatiotemporal alignment processing, thus overcoming the limitation of traditional methods that ignore the interaction of multiple physics fields. Based on the obtained multi-dimensional feature vectors, multi-physics coupling effects are quantitatively modeled to obtain a coupling feature matrix that includes the corrected electro-thermal coupling coefficient, thermo-mechanical coupling coefficient, normalized electro-mechanical coupling coefficient, maximum temperature gradient, and strain dominated by mechanical stress. This accurately characterizes the complex relationships between various physics fields under extreme conditions. By using a pre-constructed coupling-sensitive failure prediction model to output the predicted failure cycle number, the accuracy of the prediction results is effectively improved. This provides strong support for ensuring the stable operation and efficient power generation of wind power systems in extreme environments, reducing system downtime and huge economic losses caused by IGBT module failures.

[0071] Example 2 As attached Figure 2As shown, the IGBT module failure prediction method provided in this embodiment 2 is basically the same in process and principle as the IGBT module failure prediction method provided in embodiment 1 above, except that: after step 3, it further includes: Step 4: Obtain the operating temperature and humidity of the IGBT module under test.

[0072] Step 5: Input the coupling feature matrix, the operating temperature and humidity of the IGBT module under test into the pre-built dynamic failure model, and output the predicted failure probability of the IGBT module under test; wherein, the pre-built dynamic failure model is a trained deep neural network model.

[0073] It should be noted that the construction process of the pre-built dynamic failure model is as follows: Step 51: Design and create the structure of a Deep Neural Network (DNN) model to obtain an initial model. The initial model's input layer has 7 nodes, corresponding to the 5 coupling features in the coupling feature matrix of Example 1 and the operating environment temperature and humidity of the IGBT module under test obtained in Step 4. The hidden layer consists of 3 fully connected layers, with 128, 64, and 32 nodes respectively, and the activation function is ReLU. The output layer is a single node, corresponding to the predicted failure probability P of the IGBT module under test. fail The activation function is Sigmoid; the loss function is binary cross-entropy.

[0074] Step 52: Obtain the multiphysics dataset of the IGBT module during historical testing, and divide it into training set, validation set and test set in a ratio of 5:2:3; The multiphysics dataset of the IGBT module includes the collector-emitter voltage, collector current, temperature time series data and strain dominated by mechanical stress of the IGBT module; Use the training set, validation set and test set to train the initial model to obtain the pre-trained model.

[0075] It should be noted that during training, a batch training mode was used, with the Adam optimizer selected and a learning rate of 0.001. An early stopping mechanism was implemented, terminating training if the validation set loss did not decrease for five consecutive rounds. Hyperparameter tuning was performed, specifically setting the number of hidden layer nodes in the grid search (64 / 128 / 256) and the Dropout rate to 0.2. The trained DNN model was required to have a prediction error of less than 8% for the probability of failure. During testing, the input data was the test set, i.e., 30% of the historical data that did not participate in training. The SHAP interpretation tool was used to quantify the impact of each feature on the predicted probability of failure P. fail The degree of contribution.

[0076] Step 53: Deploy the pre-trained model within the converter's control system, and perform elastic incremental learning and online updates on the deployed model to obtain a trained deep neural network model; the process of elastic incremental learning and online updates is as follows: Step 531: Online acquisition of collector-emitter voltage including IGBT module collector current Temperature time series data and strain dominated by mechanical stress The data stream yields an incremental data set D. new Data snapshots are stored every 5 minutes and aggregated into batches every 24 hours.

[0077] Step 532, Coupled-Aware Elastic Update. Specifically, first, calculate the coupled correction gradient; second, for the incremental dataset D... new In the current model parameters Under the condition of the parameter importance matrix Ω, elastic constraint updates are performed to obtain a trained deep neural network model, which is then used as a pre-constructed dynamic failure model.

[0078] The process of calculating the coupled correction gradient is as follows:

[0079] in, The gradient before coupling correction; This is the gradient after coupling correction.

[0080] The process of updating elastic constraints is as follows:

[0081] in, For the updated model parameters, These are built-in parameters for elastic constraints. These are the model parameters before the update. The elastic coefficient is 0.1.

[0082] Step 6: Compare the predicted failure probability of the IGBT module under test with a preset warning threshold, and output a failure warning signal for the IGBT module under test. Preferably, the preset threshold includes a warning threshold. With warning threshold Warning threshold With warning threshold It can be dynamically adjusted and can be customized according to user requirements.

[0083] Specifically, the predicted failure probability of the IGBT module under test is compared with the early warning threshold. or warning threshold The comparisons are performed to obtain the probability comparison results; where, if the probability comparison result is the probability of prediction failure P for three consecutive predictions... fail All are greater than or equal to the warning threshold If the probability comparison result is the predicted failure probability P, then a "maintain immediately" signal will be generated and sent; fail All are greater than or equal to the warning threshold If the probability comparison result is otherwise, a "warning monitoring" signal will be generated and sent; otherwise, a "normal" signal will be generated and sent.

[0084] It should be noted that the predicted failure probability of the IGBT module under test can also be used for maintenance decision recommendations. Specifically, the SHAP interpretation tool is used to quantify the contribution of each feature to the predicted failure probability of the IGBT module under test, and based on the specific contribution, it is recommended to optimize the drive resistor Rg or reduce the load current of the IGBT module under test.

[0085] The IGBT module failure prediction method described in Embodiment 2 inputs the coupling feature matrix and the obtained operating temperature and humidity of the IGBT module under test into a pre-constructed dynamic failure model, which can accurately output the predicted failure probability. The predicted failure probability is compared with the warning threshold to output a failure warning signal, which can promptly remind maintenance, avoid wind power generation system shutdown due to IGBT module failure, reduce economic losses, and ensure stable and efficient system operation. The pre-constructed dynamic failure model, through careful structure design, reasonable division of dataset training, and elastic incremental learning and online updates, can consider multiple factors and continuously optimize to adapt to actual changes, thereby improving prediction reliability.

[0086] Example 3 As attached Figure 3 As shown in the figure, this embodiment 3 provides an IGBT module failure prediction system, including a data acquisition and feature extraction module, a coupling modeling module, and a coupling-sensitive failure prediction module.

[0087] The data acquisition and feature extraction module is used to acquire electrical stress data, thermal stress data and mechanical stress data of the IGBT module under test, and perform spatiotemporal alignment processing to obtain the multidimensional feature vector of the IGBT module under test.

[0088] The coupling modeling module is used to perform quantitative modeling of multi-physics coupling effects based on the multi-dimensional feature vector of the IGBT module under test, and obtain the coupling feature matrix. The coupling feature matrix includes the corrected electro-thermal coupling coefficient, the thermo-mechanical coupling coefficient, the normalized electro-mechanical coupling coefficient, the maximum temperature gradient, and the strain dominated by mechanical stress.

[0089] The coupling-sensitive failure prediction module is used to input the coupling feature matrix into the pre-built coupling-sensitive failure prediction model and output the predicted failure cycle number of the IGBT module under test; wherein, the pre-built coupling-sensitive failure prediction model is a random forest regression model.

[0090] Example 4 As attached Figure 4 As shown, the IGBT module failure prediction system provided in this embodiment 4 is basically the same as the IGBT module failure prediction system provided in embodiment 3 above. The difference is that it also includes an environmental data acquisition module, a failure probability prediction module and a failure early warning module.

[0091] The environmental data acquisition module is used to obtain the operating temperature and humidity of the IGBT module under test.

[0092] The failure probability prediction module is used to input the coupling feature matrix, the operating temperature and humidity of the IGBT module under test into a pre-built dynamic failure model, and output the predicted failure probability of the IGBT module under test; wherein, the pre-built dynamic failure model is a trained deep neural network model.

[0093] The failure warning module is used to compare the predicted failure probability of the IGBT module under test with the preset warning threshold and output a failure warning signal for the IGBT module under test.

[0094] It is worth noting that the specific functions of each module in the IGBT module failure prediction system described in Embodiments 3 and 4 are detailed in the corresponding content of Embodiments 1 or 2 above, and will not be repeated here.

[0095] The IGBT module failure prediction method described in this invention quantifies and models the multi-physics coupling effect of the multi-dimensional feature vector of the IGBT module under test. It utilizes a pre-constructed coupling-sensitive failure prediction model and a pre-constructed dynamic failure model, which have clear physical meanings and effectively improve failure prediction accuracy. This method can effectively realize failure analysis of IGBT modules under the combined effect of multiple factors, providing a scientific basis for IGBT module selection, life prediction, and inverter system optimization. The pre-constructed dynamic failure model supports online model updates, adapting to dynamic changes in actual operating conditions.

[0096] The above embodiments are merely one of the implementation methods for achieving the technical solution of the present invention. The scope of protection claimed by the present invention is not limited to this embodiment, but also includes any variations, substitutions and other implementation methods that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention.

Claims

1. A method for predicting IGBT module failure, characterized in that, include: The electrical stress data, thermal stress data, and mechanical stress data of the IGBT module under test are acquired and spatiotemporally aligned to obtain the multidimensional feature vector of the IGBT module under test. Based on the multidimensional feature vector of the IGBT module under test, a quantitative model of multiphysics coupling effect is performed to obtain the coupling feature matrix. The coupling feature matrix includes the corrected electro-thermal coupling coefficient, the thermo-mechanical coupling coefficient, the normalized electro-mechanical coupling coefficient, the maximum temperature gradient, and the strain dominated by mechanical stress. The coupling feature matrix is ​​input into a pre-built coupling-sensitive failure prediction model, and the predicted failure cycle number of the IGBT module under test is output. The pre-built coupling-sensitive failure prediction model is a random forest regression model.

2. The IGBT module failure prediction method according to claim 1, characterized in that, Also includes: Obtain the operating ambient temperature and humidity of the IGBT module under test; The coupling feature matrix, the operating temperature and humidity of the IGBT module under test are input into the pre-built dynamic failure model, and the predicted failure probability of the IGBT module under test is output; wherein, the pre-built dynamic failure model is a trained deep neural network model. The predicted failure probability of the IGBT module under test is compared with the preset warning threshold, and the failure warning signal of the IGBT module under test is output.

3. The IGBT module failure prediction method according to claim 1, characterized in that, The electrical stress data of the IGBT module under test includes switching transient waveform data, switching loss characteristics, and current change rate characteristics; The thermal stress data of the IGBT module under test includes junction temperature distribution data, heat sink temperature, and maximum temperature gradient; The mechanical stress data of the IGBT module under test includes microstrain data, principal vibration frequency, and strain dominated by mechanical stress.

4. The IGBT module failure prediction method according to claim 3, characterized in that, The modeling process for the corrected electro-thermal coupling coefficient is as follows: in, This is the corrected electro-thermal coupling coefficient; The defined electro-thermal coupling coefficient; The predetermined ambient temperature; The maximum temperature gradient; For activation losses; To shut down losses; This represents the average value of the test current.

5. The IGBT module failure prediction method according to claim 3, characterized in that, The modeling process for the thermo-mechanical coupling coefficient is as follows: in, Thermo-mechanical coupling coefficient; Strain dominated by mechanical stress; The dominant vibration frequency; Characterized by thermal stress; For reference frequency; This represents the maximum value of the temperature time series matrix; This is the temperature timing data for the radiator.

6. The IGBT module failure prediction method according to claim 3, characterized in that, The modeling process for the normalized electromechanical coupling coefficient is as follows: in, The normalized electromechanical coupling coefficient; The defined electromechanical coupling coefficient; The current rise rate; This is the short-circuit electromotive force.

7. The IGBT module failure prediction method according to claim 3, characterized in that, The calculation process for the maximum temperature gradient is as follows: in, The maximum temperature gradient; This is the temperature time series matrix; Corresponding to ( In two-dimensional coordinates The temperature value at time t, where k is the resolution of the spatial temperature distribution field.

8. The IGBT module failure prediction method according to claim 3, characterized in that, The calculation process for strain dominated by mechanical stress is as follows: in, Strain dominated by mechanical stress; This is a time series of strain caused by thermal expansion; The coefficient of thermal expansion of the substrate material. For reference temperature; This is the temperature time series of the radiator.

9. An IGBT module failure prediction system, characterized in that, include: The data acquisition and feature extraction module is used to acquire the electrical stress data, thermal stress data and mechanical stress data of the IGBT module under test, and perform spatiotemporal alignment processing to obtain the multidimensional feature vector of the IGBT module under test. The coupling modeling module is used to quantitatively model the multi-physics coupling effect based on the multi-dimensional feature vector of the IGBT module under test, and obtain the coupling feature matrix. The coupling feature matrix includes the corrected electro-thermal coupling coefficient, the thermo-mechanical coupling coefficient, the normalized electro-mechanical coupling coefficient, the maximum temperature gradient, and the strain dominated by mechanical stress. The coupling-sensitive failure prediction module is used to input the coupling feature matrix into the pre-built coupling-sensitive failure prediction model and output the predicted failure cycle number of the IGBT module under test; wherein, the pre-built coupling-sensitive failure prediction model is a random forest regression model.

10. The IGBT module failure prediction system according to claim 9, characterized in that, Also includes: The environmental data acquisition module is used to acquire the operating temperature and humidity of the IGBT module under test. The failure probability prediction module is used to input the coupling feature matrix, the operating temperature and humidity of the IGBT module under test into a pre-built dynamic failure model, and output the predicted failure probability of the IGBT module under test; wherein, the pre-built dynamic failure model is a trained deep neural network model. The failure warning module is used to compare the predicted failure probability of the IGBT module under test with the preset warning threshold and output a failure warning signal for the IGBT module under test.

Images