An IGBT module performance test method based on multi-parameter collaborative analysis

Abstract

The application discloses an IGBT module performance test method based on multi-parameter collaborative analysis, comprising the following steps: S1: collecting test data of the IGBT module, wherein the test data comprises an electrical parameter set, a thermal parameter set and a vibration parameter set; S2: performing a collaborative feature extraction operation based on the parameter correlation relationship among the electrical parameter set, the thermal parameter set and the vibration parameter set to generate a joint feature vector; S3: processing the joint feature vector by using a feature mapping algorithm to output performance analysis results containing a switching loss rate, a junction temperature rise slope and a vibration power spectral density; and S4: comparing the performance analysis results with parameter thresholds in a preset performance standard library, determining the performance grade of the IGBT module and generating a test report. The IGBT module is comprehensively evaluated and objectively determined in complex working conditions, and the test precision and the reliability prediction capability are significantly improved.

Description

Technical Field

[0001] This invention relates to the field of performance testing technology, and in particular to a performance testing method for IGBT modules based on multi-parameter collaborative analysis. Background Technology

[0002] Currently, performance testing of IGBT (Insulated Gate Bipolar Transistor) modules in the industrial sector commonly employs single-parameter evaluation methods, which involve testing electrical, thermal, or mechanical characteristics in isolation. While this method is simple to operate, it fails to reflect the interactive effects between parameters, leading to significant discrepancies between test results and actual application scenarios. For example, traditional methods typically measure switching characteristics under constant temperature conditions, ignoring the impact of temperature fluctuations on electrical performance during actual operation; or they focus only on static parameters while neglecting performance under dynamic operating conditions. Furthermore, existing technologies largely rely on manual experience to judge test results, lacking a standardized evaluation system. This makes objective comparisons between IGBT modules from different batches and manufacturers difficult, increasing the difficulty and risk of power electronic system design and selection.

[0003] The limitations of traditional testing methods are particularly pronounced in high-reliability applications, such as new energy vehicle drive systems and high-speed railway traction systems, where the overall performance requirements for IGBT modules are extremely high. Due to the lack of multi-parameter collaborative analysis capabilities, existing technologies struggle to accurately assess the performance boundaries of IGBTs under complex operating conditions and cannot effectively predict their reliability levels in practical applications. Furthermore, parameter extraction and analysis in traditional testing processes are largely manual, resulting in low efficiency and subject to subjective bias, which is detrimental to rapid screening and quality control on the production line.

[0004] These technical deficiencies severely restrict the development of IGBT modules in high-end applications, and there is an urgent need for a comprehensive performance testing method that can achieve multi-parameter collaborative analysis and objective evaluation. Summary of the Invention

[0005] This invention provides a performance testing method for IGBT modules based on multi-parameter collaborative analysis to solve the aforementioned problems in the prior art.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A performance testing method for IGBT modules based on multi-parameter collaborative analysis includes:

[0008] S1: Collect test data of the IGBT module, including electrical parameter set, thermal parameter set and vibration parameter set;

[0009] S2: Based on the parameter correlation between the electrical parameter set, thermal parameter set and vibration parameter set, perform collaborative feature extraction operation to generate joint feature vector;

[0010] S3: Use the feature mapping algorithm to process the joint feature vector and output the performance analysis results including switching loss rate, junction temperature rise slope and vibration power spectral density.

[0011] S4: Compare the performance analysis results with the parameter thresholds in the preset performance standard library to determine the performance level of the IGBT module and generate a test report.

[0012] Step S1 includes:

[0013] S11: Start the IGBT module performance test system, apply an electrothermal load to the IGBT module based on preset voltage and current parameters, and simultaneously collect the raw data stream;

[0014] S12: Perform signal separation processing on the raw data stream to extract electrical raw data, thermal raw data and vibration raw data;

[0015] S13: Call the pre-stored electrical calibration coefficient matrix to perform linear compensation on the original electrical data, call the thermal calibration coefficient matrix to perform nonlinear correction on the original thermal data, call the vibration calibration coefficient matrix to perform frequency domain calibration on the original vibration data, and generate the calibrated parameter set.

[0016] Step S2 includes:

[0017] S21: Analyze the electrical parameter set to obtain the electrical feature subset, analyze the thermal parameter set to obtain the thermal feature subset, and analyze the vibration parameter set to obtain the vibration feature subset;

[0018] S22: Calculate the correlation coefficient between the electrical feature subset and the thermal feature subset to construct the first correlation matrix; calculate the mutual information entropy between the electrical feature subset and the vibration feature subset to construct the second correlation matrix; calculate the covariance between the thermal feature subset and the vibration feature subset to construct the third correlation matrix.

[0019] S23: Merge the first correlation matrix, the second correlation matrix, and the third correlation matrix to generate a joint feature vector.

[0020] Step S4 includes:

[0021] S41: Extract switching loss rate, junction temperature rise slope and vibration power spectral density from the performance analysis results;

[0022] S42: Compare the switching loss rate, junction temperature rise slope, and vibration power spectral density with the corresponding parameter threshold ranges in the preset performance standard library;

[0023] S43: When all switching loss rates, junction temperature rise slopes, and vibration power spectral density are within the corresponding parameter threshold ranges, the result is marked as qualified; otherwise, the fault diagnosis procedure is executed.

[0024] Step S13 includes:

[0025] S131: Perform matrix operations on the electrical calibration coefficient matrix and the original electrical data to achieve linear compensation;

[0026] S132: Perform nonlinear function calculations on the thermal calibration coefficient matrix and the original thermal data to achieve temperature drift correction;

[0027] S133: Perform a convolution operation between the vibration calibration coefficient matrix and the spectrum of the original vibration data to achieve frequency domain amplitude calibration.

[0028] Step S21 includes:

[0029] S211: Extract rise time, turn-off overshoot voltage and on-state voltage drop from electrical parameter set;

[0030] S212: Extract junction temperature gradient distribution and thermal resistance curve slope from thermal parameter set;

[0031] S213: Extract the resonant frequency, root mean square value of vibration acceleration, and spectral energy distribution from the vibration parameter set.

[0032] Step S23 includes:

[0033] S231: Perform eigenvalue decomposition on the first, second, and third incidence matrices respectively;

[0034] S232: Select the eigenvector corresponding to the largest eigenvalue from the eigenvalue decomposition results of each matrix as the dominant eigencomponent;

[0035] S233: Merge the dominant feature components according to predetermined weights to generate a joint feature vector.

[0036] Step S11 further includes:

[0037] S111: Collect raw vibration data;

[0038] S112: Calculate the root mean square value of vibration acceleration from the raw vibration data;

[0039] S113: Determine whether the root mean square value of vibration acceleration exceeds a preset threshold;

[0040] S114: When the root mean square value of vibration acceleration exceeds the preset threshold, increase the sampling rate of electrical parameters.

[0041] The fault diagnosis procedure in step S43 includes:

[0042] S431: Analyze the high-frequency components of the second correlation matrix. When the mutation amount of the high-frequency component exceeds the preset critical value, it is marked as a wire bond failure.

[0043] S432: Analyze the low-frequency energy ratio of the third correlation matrix. When the low-frequency energy ratio exceeds the preset threshold, it is marked as a heat dissipation substrate delamination fault.

[0044] S433: Generate a fault report that includes the fault type code and location coordinates.

[0045] Step S22 further includes:

[0046] S221: Analyze the frequency domain energy distribution characteristics of the vibration feature subset;

[0047] S222: Determine the optimal sampling frequency based on the frequency domain energy distribution characteristics;

[0048] S223: Apply the optimized sampling frequency to the subsequent acquisition of electrical feature subsets;

[0049] S224: Update the electrical feature subset based on the optimized sampling frequency.

[0050] Compared with the prior art, the present invention has the following advantages:

[0051] A performance testing method for IGBT modules based on multi-parameter collaborative analysis includes: S1: collecting test data of the IGBT module, including electrical parameter sets, thermal parameter sets, and vibration parameter sets; S2: performing collaborative feature extraction based on the parameter correlation between the electrical parameter sets, thermal parameter sets, and vibration parameter sets to generate a joint feature vector; S3: processing the joint feature vector using a feature mapping algorithm to output performance analysis results including switching loss rate, junction temperature rise slope, and vibration power spectral density; S4: comparing the performance analysis results with parameter thresholds in a preset performance standard library to determine the performance level of the IGBT module and generate a test report. This method achieves comprehensive performance evaluation and objective level determination of IGBT modules under complex operating conditions, significantly improving test accuracy and reliability prediction capabilities.

[0052] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention.

[0053] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0054] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0055] Figure 1 This is a flowchart of a performance testing method for an IGBT module based on multi-parameter collaborative analysis in an embodiment of the present invention;

[0056] Figure 2 This is a flowchart illustrating the generation of joint feature vectors in an embodiment of the present invention. Detailed Implementation

[0057] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0058] The embodiments of the present invention provide, as follows Figure 1 As shown, a performance testing method for IGBT modules based on multi-parameter collaborative analysis includes:

[0059] S1: Collect test data of the IGBT module, including electrical parameter set, thermal parameter set and vibration parameter set;

[0060] S2: Based on the parameter correlation between the electrical parameter set, thermal parameter set and vibration parameter set, perform collaborative feature extraction operation to generate joint feature vector;

[0061] S3: Use the feature mapping algorithm to process the joint feature vector and output the performance analysis results including switching loss rate, junction temperature rise slope and vibration power spectral density.

[0062] S4: Compare the performance analysis results with the parameter thresholds in the preset performance standard library to determine the performance level of the IGBT module and generate a test report.

[0063] The working principle and beneficial effects of the above technical solution are as follows: In S1, the testing instrument monitors various signals generated by the IGBT module in real time during operation through a multi-channel data acquisition device. Specifically, during module operation, voltage and current waveforms during the switching process are acquired using voltage and current probes to form an electrical parameter set; simultaneously, an infrared thermal imager and a temperature sensor are used to record the surface temperature distribution of the module to form a thermal parameter set; and an accelerometer is used to capture the vibration state of the module during operation to form a vibration parameter set. These parameters are transmitted to the data processing unit through a dedicated interface for subsequent analysis.

[0064] In S2, the data processing unit first aligns the three types of parameters according to time labels and removes obvious outliers. Then, it extracts feature points from the voltage rising edge and current falling edge of the electrical parameters, calculates the peak point of the temperature gradient change rate of the thermal parameters, and extracts the dominant frequency component from the spectral distribution of the vibration parameters. Next, it establishes a time-series correlation table between these feature points, analyzes the response characteristics of the thermal and vibration parameters when the switching state changes, and combines these correlation characteristics into a joint feature vector that comprehensively characterizes the module's performance using a weighted fusion algorithm.

[0065] In S3, the system inputs the joint feature vector into a pre-trained feature mapping network for processing. This network maps the original feature space to a performance index space through multi-layer nonlinear transformations. Specifically, it uses piecewise linear transformations to process electrical features to obtain the switching loss rate; employs exponential fitting to process the temperature curve to extract the junction temperature rise slope; and uses fast Fourier transform to process vibration data to calculate the power spectral density. These calculation results are combined to form the final performance analysis report, visually displaying the module's key performance indicators.

[0066] In S4, the system reads the built-in performance standard library and extracts the threshold parameter table corresponding to the IGBT model being tested. Then, it compares the performance indicators obtained in S3 with the standard thresholds one by one: first, it checks whether the switching loss rate is lower than the maximum allowable value; then, it verifies whether the junction temperature rise slope is within a safe range; and finally, it confirms whether the vibration power spectral density meets the stability requirements. Based on the comparison results, the system automatically evaluates the module's performance level (e.g., excellent, good, qualified, or unqualified) and generates a test report containing detailed test data and evaluation conclusions.

[0067] In another embodiment, step S1 includes:

[0068] S11: Start the IGBT module performance test system, apply an electrothermal load to the IGBT module based on preset voltage and current parameters, and simultaneously collect the raw data stream;

[0069] S12: Perform signal separation processing on the raw data stream to extract electrical raw data, thermal raw data and vibration raw data;

[0070] S13: Call the pre-stored electrical calibration coefficient matrix to perform linear compensation on the original electrical data, call the thermal calibration coefficient matrix to perform nonlinear correction on the original thermal data, call the vibration calibration coefficient matrix to perform frequency domain calibration on the original vibration data, and generate the calibrated parameter set.

[0071] The working principle and beneficial effects of the above technical solution are as follows: In S11, the operator starts the IGBT module performance testing system through the control panel, sets the test parameters, and clicks the "Start Test" button. The system then applies an electrothermal load to the IGBT module under test through the power drive circuit according to the preset voltage and current parameter values ​​(e.g., gate voltage ±15V, collector current 100A). During this process, multiple data acquisition channels work synchronously, recording various electrical signals, temperature changes, and mechanical vibrations generated by the module under load at a sampling rate of up to 1MHz, forming a raw data stream containing multiple physical quantities.

[0072] In S12, after receiving the mixed raw data stream, the data processing unit first separates it using channel identification technology. Specifically, the data stream is first analyzed in the time domain to identify the sampling timestamps of different signals; then, based on predefined signal feature templates, voltage and current waveforms belonging to electrical parameters are extracted to form electrical raw data; data sequences characterizing temperature changes are separated to form thermal raw data; and acceleration signals containing vibration characteristics are extracted to form vibration raw data. In this way, the originally mixed data stream is clearly divided into three categories of raw data.

[0073] In step S13, the system calibrates three types of raw data separately. First, it reads a pre-stored electrical calibration coefficient matrix from the system configuration file. This matrix contains parameters such as voltage scaling factors and current sensor sensitivity. The system applies these coefficients to the electrical raw data for linear compensation, correcting measurement errors. Next, it calls the thermal calibration coefficient matrix, which contains correction values ​​at different temperature points. Based on this, the system performs piecewise function correction on the thermal raw data to eliminate nonlinear errors in the sensor. Finally, it loads the vibration calibration coefficient matrix and performs frequency domain calibration on the vibration raw data based on the response characteristics of the accelerometer at different frequency bands, ensuring the accuracy of the vibration signal across the entire frequency range.

[0074] In another embodiment, such as Figure 2 As shown, step S2 includes:

[0075] S21: Analyze the electrical parameter set to obtain the electrical feature subset, analyze the thermal parameter set to obtain the thermal feature subset, and analyze the vibration parameter set to obtain the vibration feature subset;

[0076] S22: Calculate the correlation coefficient between the electrical feature subset and the thermal feature subset to construct the first correlation matrix; calculate the mutual information entropy between the electrical feature subset and the vibration feature subset to construct the second correlation matrix; calculate the covariance between the thermal feature subset and the vibration feature subset to construct the third correlation matrix.

[0077] S23: Merge the first correlation matrix, the second correlation matrix, and the third correlation matrix to generate a joint feature vector.

[0078] The working principle and beneficial effects of the above technical solution are as follows: In S21, the processor analyzes the three sets of calibrated parameters and extracts their respective feature information. For the electrical parameter set, the system calculates indicators such as voltage rise time, current fall time, turn-on energy, and turn-off energy during the switching process, forming an electrical feature subset; for the thermal parameter set, the system extracts indicators such as junction temperature change rate, temperature distribution uniformity, and heat diffusion time constant, forming a thermal feature subset; for the vibration parameter set, the system analyzes indicators such as vibration frequency distribution, amplitude characteristics, and harmonic components, forming a vibration feature subset. These three feature subsets characterize the working characteristics of the IGBT module from different perspectives.

[0079] In S22, the system performs a quantitative analysis of the interrelationships among the three feature subsets. First, the program aligns the switching time series from the electrical feature subset with the temperature response data from the thermal feature subset, calculates the Pearson correlation coefficient between each pair of parameters, and arranges these coefficients into a matrix to form the first correlation matrix reflecting the electrothermal coupling relationship. Then, the system takes the current change rate data from the electrical feature subset and the vibration response data from the vibration feature subset, calculates their mutual information entropy, and quantifies the information transfer relationship between electrical switching and mechanical response; these entropy values ​​form the second correlation matrix. Finally, the system pairs the temperature rise data from the thermal feature subset with the spectral variation data from the vibration feature subset, calculates the covariance of each pair of parameters, and forms the third correlation matrix characterizing the correlation between thermal state and mechanical vibration.

[0080] In S23, the system merges the three correlation matrices into a unified feature representation. Specifically, each correlation matrix is ​​first normalized to make data from different physical dimensions comparable; then, principal eigenvectors are extracted from each matrix to reduce data redundancy; finally, these principal eigenvectors are combined with certain weights to form a joint feature vector that comprehensively reflects the multi-physics coupling characteristics of the IGBT module. This vector contains comprehensive information about the module's electrical performance, thermal performance, and mechanical characteristics, providing a foundation for subsequent performance analysis.

[0081] In another embodiment, step S4 includes:

[0082] S41: Extract switching loss rate, junction temperature rise slope and vibration power spectral density from the performance analysis results;

[0083] S42: Compare the switching loss rate, junction temperature rise slope, and vibration power spectral density with the corresponding parameter threshold ranges in the preset performance standard library;

[0084] S43: When all switching loss rates, junction temperature rise slopes, and vibration power spectral density are within the corresponding parameter threshold ranges, the result is marked as qualified; otherwise, the fault diagnosis procedure is executed.

[0085] The working principle and beneficial effects of the above technical solution are as follows: In S41, the processor identifies the energy loss region in the switching waveform and calculates the switching loss rate through integration. This parameter directly reflects the energy conversion efficiency of the IGBT. Subsequently, the system performs linear fitting on the temperature curve and extracts the junction temperature rise slope, which characterizes the chip's heat dissipation performance and thermal stability. Finally, the system performs a fast Fourier transform on the vibration signal to calculate the vibration power spectral density, quantifying the module's mechanical vibration characteristics and structural stability.

[0086] In S42, the system performs a multi-dimensional parameter comparison operation. The processor first accesses the built-in performance standard library to extract the parameter threshold ranges corresponding to the IGBT model under test. Subsequently, the system compares the three extracted core indicators with the corresponding threshold ranges one by one. The comparison process uses interval matching technology; the system not only determines whether the parameter is within the qualified range but also calculates its deviation from the center of the range, quantifying the performance margin. For parameters in critical states, the system adds auxiliary criteria for secondary evaluation.

[0087] In S43, the system performs a comprehensive judgment based on the comparison results. When all three core indicators fall within their respective qualified threshold ranges, the system marks the module under test as qualified and further subdivides it into three sub-levels: excellent, good, or basically qualified, based on the performance margin. If any indicator exceeds the threshold range, the system immediately initiates a fault diagnosis program, using feature pattern recognition to locate possible defect types, such as abnormal gate drive, chip bonding layer damage, or loose packaging structure. Based on the diagnostic results, the system generates a detailed report containing the specific defect location and severity.

[0088] In another embodiment, step S13 includes:

[0089] S131: Perform matrix operations on the electrical calibration coefficient matrix and the original electrical data to achieve linear compensation;

[0090] S132: Perform nonlinear function calculations on the thermal calibration coefficient matrix and the original thermal data to achieve temperature drift correction;

[0091] S133: Perform a convolution operation between the vibration calibration coefficient matrix and the spectrum of the original vibration data to achieve frequency domain amplitude calibration.

[0092] The working principle and beneficial effects of the above technical solution are as follows: In step S131, the system performs linear compensation on the raw electrical data. The compensation process first constructs a calibration coefficient matrix containing gain coefficients and bias terms, which is obtained through high-precision reference source measurements. Then, the system performs matrix multiplication on the raw electrical data and the calibration coefficient matrix, while simultaneously adding the bias correction value. This linear compensation method effectively eliminates system errors in the acquisition channel and improves the absolute accuracy of voltage and current measurements.

[0093] In step S132, the system performs nonlinear function calculations on the raw thermal data. Thermal data calibration considers the sensor's nonlinear characteristics and the influence of ambient temperature, employing a piecewise polynomial fitting method. The system first selects the corresponding calibration coefficient set based on the ambient temperature, and then substitutes the raw thermal data into the nonlinear calibration function for calculation. This calibration method effectively compensates for the response differences of the temperature sensor in different temperature zones, ensuring the consistency of temperature measurements across the entire measurement range.

[0094] In S133, the system performs frequency domain calibration on the raw vibration data. The calibration process first performs a Fast Fourier Transform (FFT) on the raw vibration signal to obtain its spectral representation. Then, it convolves the pre-calibrated frequency response calibration coefficient matrix with the vibration signal's spectrum. This convolution operation essentially applies different correction factors to different frequency components, compensating for inherent frequency response characteristics of the sensor and resonances caused by installation conditions. Finally, an Inverse Fourier Transform (IFT) is used to convert the calibrated spectrum back to the time domain, yielding the calibrated vibration data.

[0095] In another embodiment, step S21 includes:

[0096] S211: Extract rise time, turn-off overshoot voltage and on-state voltage drop from electrical parameter set;

[0097] S212: Extract junction temperature gradient distribution and thermal resistance curve slope from thermal parameter set;

[0098] S213: Extract the resonant frequency, root mean square value of vibration acceleration, and spectral energy distribution from the vibration parameter set.

[0099] The working principle and beneficial effects of the above technical solution are as follows: In S211, the system retrieves specific technical indicators from the electrical parameter data records. The rise time is calculated by detecting the voltage change from low to high and measuring the time interval between 10% and 90% of the voltage level. The turn-off overshoot voltage is obtained by capturing the maximum voltage value at the moment the device turns off, subtracting the steady-state voltage, and the difference is the overshoot. The on-state voltage drop is calculated by taking the average voltage over a stable range after the IGBT is fully turned on. These electrical characteristic indicators reflect the switching speed and conduction losses of the device.

[0100] In S212, thermal parameter extraction involves examining the temperature distribution. The junction temperature gradient distribution is obtained by processing temperature field images captured by a thermal imager, calculating the rate of temperature change between adjacent points, and forming a temperature gradient map. The slope of the thermal resistance curve is calculated by correlating power changes with corresponding temperature rise changes and fitting a straight line to these data points; the slope of the line represents the thermal resistance characteristic value. This part of the work must be performed in a relatively stable temperature environment to avoid external interference.

[0101] In S213, the resonant frequency is the frequency point where the energy is most concentrated after the vibration signal has been transformed. The root mean square value is calculated by squaring the vibration waveform data, integrating, and then taking the square root, reflecting the vibration intensity. The spectral energy distribution divides the entire frequency range into several segments and calculates the proportion of energy in each segment to the total energy. These vibration characteristics can reflect the state of the internal mechanical structure of the device.

[0102] In another embodiment, step S23 includes:

[0103] S231: Perform eigenvalue decomposition on the first, second, and third incidence matrices respectively;

[0104] S232: Select the eigenvector corresponding to the largest eigenvalue from the eigenvalue decomposition results of each matrix as the dominant eigencomponent;

[0105] S233: Merge the dominant feature components according to predetermined weights to generate a joint feature vector.

[0106] The working principle and beneficial effects of the above technical solution are as follows: In S231, the matrix operation program processes three types of correlated data. The operator starts the matrix analysis tool and imports the three correlated matrices sequentially; the program automatically calculates the characteristic equations of each matrix, solves for all eigenvalues ​​and corresponding eigenvectors; the output includes a list of eigenvalues ​​and corresponding eigenvector groups for each matrix. This step transforms complex correlated data into a quantifiable and analyzable characteristic form.

[0107] In S232, the feature filtering module performs principal feature extraction. The program scans the eigenvalue list of the first correlation matrix, finds the item with the largest value and records its position label; it reads the eigenvector corresponding to that position and saves it as the electrothermal correlation principal feature; similarly, it extracts the electro-vibration correlation principal feature from the second correlation matrix and the thermal vibration correlation principal feature from the third correlation matrix. These three vectors contain the most significant physical correlation information.

[0108] In S233, the feature fusion unit performs a vector merging operation. Technicians read preset weight values ​​from the system configuration items and specify importance coefficients for the three correlations: electrothermal, electro-vibration, and thermal vibration. The program multiplies the three main feature vectors by their corresponding weights and then concatenates them into a composite vector. This joint feature vector simultaneously contains the core coupling characteristics between the three types of physical quantities, facilitating subsequent overall evaluation.

[0109] In another embodiment, step S11 further includes:

[0110] S111: Collect raw vibration data;

[0111] S112: Calculate the root mean square value of vibration acceleration from the raw vibration data;

[0112] S113: Determine whether the root mean square value of vibration acceleration exceeds a preset threshold;

[0113] S114: When the root mean square value of vibration acceleration exceeds the preset threshold, increase the sampling rate of electrical parameters.

[0114] The working principle and beneficial effects of the above technical solution are as follows: Step S111 is to acquire the raw vibration signal. A triaxial accelerometer is installed on the experimental platform and attached to several key locations on the surface of the IGBT housing. The sensor signal is converted into a voltage signal by a charge amplifier, then passed through a low-pass filter to remove high-frequency noise, and finally recorded by a data acquisition card. During the sampling process, attention should be paid to ground vibration isolation to prevent external interference from affecting the measurement results.

[0115] Step S112 involves processing this vibration data. The acquisition software divides the vibration data over a long period into several windows, each containing a certain number of sampling points. For each data point within a window, the square value is first calculated, summed, divided by the number of points, and then the square root is taken to obtain the root mean square value of that window. Finally, the results from all windows are averaged to serve as an indicator of the overall vibration intensity.

[0116] Step S113 determines whether the vibration is abnormal. The program compares the root mean square value calculated in the previous step with a pre-set threshold. This threshold is statistically derived from vibration data of a large number of normally operating IGBT modules and is generally set to about twice the normal value. A comparator with hysteresis is used during the comparison to avoid frequent changes in the judgment result when the vibration fluctuates near the critical value.

[0117] Step S114 adjusts the sampling strategy based on vibration. When vibration exceeds the warning value, it indicates a possible abnormality. At this point, the control program automatically increases the sampling frequency of electrical parameters, raising the original sampling rate by 3-5 times. The increased sampling rate captures more detailed changes in electrical parameters, especially the waveform details during switching, which helps identify the cause of increased vibration.

[0118] In another embodiment, the fault diagnosis procedure in step S43 includes:

[0119] S431: Analyze the high-frequency components of the second correlation matrix. When the mutation amount of the high-frequency component exceeds the preset critical value, it is marked as a wire bond failure.

[0120] S432: Analyze the low-frequency energy ratio of the third correlation matrix. When the low-frequency energy ratio exceeds the preset threshold, it is marked as a heat dissipation substrate delamination fault.

[0121] S433: Generate a fault report that includes the fault type code and location coordinates.

[0122] The working principle and beneficial effects of the above technical solution are as follows: In step S431, the test system performs frequency band decomposition on the second correlation matrix, extracts components above 400kHz, and constructs a high-frequency feature spectrum. By comparing the Euclidean distance between the current test sample and the standard template, the amplitude of the high-frequency component mutation is calculated. When the mutation amount exceeds a set critical value, the wire bonding region is located using a pattern recognition algorithm, confirming a bond detachment fault.

[0123] In step S432, the system transforms the third correlation matrix to the frequency domain and calculates the proportion of energy in the 0-200Hz low-frequency band in the total. The calculation process first normalizes the matrix and then applies discrete cosine transform to obtain the spectral distribution. If the proportion of low frequencies exceeds a preset standard, an abnormal heat dissipation conduction path is inferred based on the thermodynamic model, and it is determined to be a substrate delamination fault.

[0124] In step S433, the fault diagnosis unit converts the fault type into a standard code according to its internal coding rules. Location is determined using triangulation, with cross-validation of multiple abnormal areas to determine the physical coordinates.

[0125] In another embodiment, step S22 further includes:

[0126] S221: Analyze the frequency domain energy distribution characteristics of the vibration feature subset;

[0127] S222: Determine the optimal sampling frequency based on the frequency domain energy distribution characteristics;

[0128] S223: Apply the optimized sampling frequency to the subsequent acquisition of electrical feature subsets;

[0129] S224: Update the electrical feature subset based on the optimized sampling frequency.

[0130] The working principle and beneficial effects of the above technical solution are as follows: Step S221 involves in-depth analysis of the vibration spectrum. The testing software first applies a windowing layer to the time-domain vibration data to reduce spectral leakage, and then uses a transformation algorithm to obtain a frequency domain representation. The program automatically identifies the frequency bands where the main energy is concentrated, calculates the energy proportion of each frequency band, and generates a spectral characteristic map. Special attention is paid to harmonic components related to the switching frequency, as these are often closely related to electrical characteristics.

[0131] Step S222 optimizes the sampling strategy based on vibration characteristics. The program first determines the highest effective frequency in the vibration signal, i.e., the frequency point where energy accumulation reaches 95%. Then, based on the spectral characteristics, it calculates the minimum sampling rate required to capture the electrical waveform, typically 2.5 to 3 times the highest effective frequency. This dynamic sampling method avoids information loss or resource waste that may result from a fixed sampling rate.

[0132] Step S223 involves actually adjusting the sampling parameters. The test master program sends a new configuration command to the data acquisition card, updates the clock division coefficient, and modifies the storage depth parameter. Simultaneously, the preamplifier gain is adjusted to ensure the signal amplitude is appropriate and does not saturate. After the sampling strategy is changed, the system will briefly pause data processing, waiting for the new settings to take effect.

[0133] Step S224 involves re-acquiring updated data. Electrical parameters are re-acquired using a new sampling rate, with particular attention to details during switching. The program extracts features from the newly acquired data, including switching time point location and waveform feature point identification, generating an updated subset of electrical features. This vibration-guided adaptive sampling method improves the ability to capture transient events, making the correlation analysis of electrical and mechanical characteristics more accurate.

[0134] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from the spirit and scope of this invention.

Claims

1. A performance testing method for IGBT modules based on multi-parameter collaborative analysis, characterized in that, include: S1: Collect test data of the IGBT module, including electrical parameter set, thermal parameter set and vibration parameter set; S2: Based on the parameter correlation between the electrical parameter set, thermal parameter set, and vibration parameter set, perform collaborative feature extraction to generate a joint feature vector; including: S21: Analyze the electrical parameter set to obtain the electrical feature subset, analyze the thermal parameter set to obtain the thermal feature subset, and analyze the vibration parameter set to obtain the vibration feature subset; S22: Calculate the correlation coefficient between the electrical feature subset and the thermal feature subset to construct the first correlation matrix; calculate the mutual information entropy between the electrical feature subset and the vibration feature subset to construct the second correlation matrix; calculate the covariance between the thermal feature subset and the vibration feature subset to construct the third correlation matrix. S23: Merge the first, second, and third correlation matrices to generate a joint feature vector; Step S23 includes: S231: Perform eigenvalue decomposition on the first, second, and third incidence matrices respectively; S232: Select the eigenvector corresponding to the largest eigenvalue from the eigenvalue decomposition results of each matrix as the dominant eigencomponent; S233: Merge the dominant feature components according to predetermined weights to generate a joint feature vector; S3: Use the feature mapping algorithm to process the joint feature vector and output the performance analysis results including switching loss rate, junction temperature rise slope and vibration power spectral density. S4: Compare the performance analysis results with the parameter thresholds in the preset performance standard library to determine the performance level of the IGBT module and generate a test report.

2. The IGBT module performance testing method based on multi-parameter collaborative analysis according to claim 1, characterized in that, Step S1 includes: S11: Start the IGBT module performance test system, apply an electrothermal load to the IGBT module based on preset voltage and current parameters, and simultaneously collect the raw data stream; S12: Perform signal separation processing on the raw data stream to extract electrical raw data, thermal raw data and vibration raw data; S13: Call the pre-stored electrical calibration coefficient matrix to perform linear compensation on the original electrical data, call the thermal calibration coefficient matrix to perform nonlinear correction on the original thermal data, call the vibration calibration coefficient matrix to perform frequency domain calibration on the original vibration data, and generate the calibrated parameter set.

3. The IGBT module performance testing method based on multi-parameter collaborative analysis according to claim 1, characterized in that, Step S4 includes: S41: Extract switching loss rate, junction temperature rise slope and vibration power spectral density from the performance analysis results; S42: Compare the switching loss rate, junction temperature rise slope, and vibration power spectral density with the corresponding parameter threshold ranges in the preset performance standard library; S43: When all switching loss rates, junction temperature rise slopes, and vibration power spectral density are within the corresponding parameter threshold ranges, the result is marked as qualified; otherwise, the fault diagnosis procedure is executed.

4. The IGBT module performance testing method based on multi-parameter collaborative analysis according to claim 2, characterized in that, Step S13 includes: S131: Perform matrix operations on the electrical calibration coefficient matrix and the original electrical data to achieve linear compensation; S132: Perform nonlinear function calculations on the thermal calibration coefficient matrix and the original thermal data to achieve temperature drift correction; S133: Perform a convolution operation between the vibration calibration coefficient matrix and the spectrum of the original vibration data to achieve frequency domain amplitude calibration.

5. The IGBT module performance testing method based on multi-parameter collaborative analysis according to claim 1, characterized in that, Step S21 includes: S211: Extract rise time, turn-off overshoot voltage and on-state voltage drop from electrical parameter set; S212: Extract junction temperature gradient distribution and thermal resistance curve slope from thermal parameter set; S213: Extract the resonant frequency, root mean square value of vibration acceleration, and spectral energy distribution from the vibration parameter set.

6. The IGBT module performance testing method based on multi-parameter collaborative analysis according to claim 2, characterized in that, Step S11 also includes: S111: Collect raw vibration data; S112: Calculate the root mean square value of vibration acceleration from the raw vibration data; S113: Determine whether the root mean square value of vibration acceleration exceeds a preset threshold; S114: When the root mean square value of vibration acceleration exceeds the preset threshold, increase the sampling rate of electrical parameters.

7. The IGBT module performance testing method based on multi-parameter collaborative analysis according to claim 4, characterized in that, The fault diagnosis procedure in step S43 includes: S431: Analyze the high-frequency components of the second correlation matrix. When the mutation amount of the high-frequency component exceeds the preset critical value, it is marked as a wire bond failure. S432: Analyze the low-frequency energy ratio of the third correlation matrix. When the low-frequency energy ratio exceeds the preset threshold, it is marked as a heat dissipation substrate delamination fault. S433: Generate a fault report that includes the fault type code and location coordinates.

8. The IGBT module performance testing method based on multi-parameter collaborative analysis according to claim 1, characterized in that, Step S22 also includes: S221: Analyze the frequency domain energy distribution characteristics of the vibration feature subset; S222: Determine the optimal sampling frequency based on the frequency domain energy distribution characteristics; S223: Apply the optimized sampling frequency to the subsequent acquisition of electrical feature subsets; S224: Update the electrical feature subset based on the optimized sampling frequency.

Images