Intelligent analysis and test method and system for IGBT (Insulated Gate Bipolar Translator) packaging structure

Abstract

The invention discloses an intelligent analyzing and testing method and system for an IGBT (Insulated Gate Bipolar Translator) packaging structure, and relates to the technical field of packaging test.A current signal set Cur is constructed, normalized features of the current signal set Cur are extracted to form a feature set Feaa, an autocorrelation curve set COR is generated through autocorrelation operation, a periodic jitter intensity coefficient Jit is calculated, and the periodic jitter intensity coefficient Jit is obtained; the cross-domain mapping from electric signal change to mechanical loosening state is realized, and the early loosening symptom of the pin can be identified through the tiny change of the periodic jitter intensity coefficient Jit only under the conventional electrical characteristic test condition. Furthermore, by comparing with a normal jitter standard value NorJ of a normal model, a deviation ratio Rat is calculated, and an anomaly judgment set Abn is formed according to a set judgment threshold Thr, so that automatic anomaly recognition and result classification are realized. And the problems of large size, long detection delay, insufficient sensitivity and the like of external excitation equipment are avoided.

Description

Technical Field

[0001] This invention relates to the field of packaging and testing technology, specifically to an intelligent analysis and testing method and system for IGBT packaging structures. Background Technology

[0002] In the field of power electronics applications, IGBTs (Insulated Gate Bipolar Transistors) are widely used in new energy vehicle drives, power grid converters, and industrial control systems due to their high voltage withstand capability and excellent switching characteristics. With the continuous increase in power density and operating frequency, the IGBT packaging structure has become a key factor determining the stability and lifespan of the device. To ensure long-term operational reliability, intelligent analysis and testing of the packaging structure has become a key focus of industry development.

[0003] In existing IGBT package structure testing, the package pins, as key components for electrical connection and mechanical support, are prone to micro-loosening after long-term vibration, thermal cycling, or mechanical shock. Current detection methods mainly rely on external vibration testing, microscopic probing, or acoustic emission techniques. These methods often require specialized equipment, involve complex testing environments, and cannot identify pin contact changes in real time while the device is powered on. Furthermore, conventional electrical testing often only focuses on macroscopic parameters such as on-resistance or leakage current, neglecting the dynamic perturbation information contained in the current waveform, thus failing to accurately identify early mechanical loosening. Faced with these detection limitations, slight changes in pin contact status are often masked at the system level, making it difficult to form effective health warnings. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides an intelligent analysis and testing method and system for IGBT packaging structures, solving the problems mentioned in the background section.

[0005] To achieve the above objectives, the present invention provides the following technical solution: an intelligent analysis and testing method for IGBT packaging structures, comprising the following steps: S1. Collect the dynamic current signal sequence of the IGBT package pins during the on / off test process and construct the current signal set Cur; S2. Perform noise filtering, amplitude normalization, and time scale alignment on the current signal set Cur to obtain the normalized signal set Nor. Divide the normalized signal set Nor into segments according to the time interval, calculate the instantaneous rate of change, peak amplitude, and period difference for each segment of the waveform, and obtain the feature set Fea used to reflect the dynamic characteristics of the waveform. S3. Perform autocorrelation function calculation on feature set Fea to generate autocorrelation curve set Cor. By analyzing the peak distribution and delay interval in autocorrelation curve set Cor, obtain the periodic jitter intensity coefficient Jit. S4. Compare the periodic jitter intensity coefficient Jit with the preset normal operation model standard value NorJ, calculate the deviation ratio Rat, and when the deviation ratio Rat exceeds the preset judgment threshold Thr, determine that the current IGBT package pin is in a loose state and generate an abnormal judgment set Abn.

[0006] Preferably, S1 includes S11; S11. During the IGBT package structure continuity test, a sampling circuit for measuring current is connected in series on the pin power path to sample the current during the continuity process of each pin. During the current sampling process, the current is sampled multiple times at fixed time intervals. The time interval is set to between one-tenth and one-twentieth of the IGBT switching frequency. During sampling, the sampling time, instantaneous current value, corresponding pin number and conduction cycle number are recorded respectively. The continuous sampling data are arranged in chronological order to form the raw current sampling set Raw. The Raw current sampling set includes continuous current variation data for each pin over multiple switching cycles.

[0007] Preferably, S1 further includes S12; S12. Group the original current sampling set Raw according to the pin number and the cycle number, sort the sampling points of each group in time order, so that the current change in each cycle is continuous without interruption, and perform time axis interpolation on the sorted data at equal intervals to correct the sampling interval error and obtain a smooth current time sequence Seq. Then, the current timing sequence Seq of the same pin under multiple cycles is uniformly organized, and all time sequences are spliced ​​together to form a complete pin current response sequence for each pin. Then, the current timing sequences Seq corresponding to all pins are summarized to generate the current signal set Cur.

[0008] Preferably, S2 includes S21; S21. Perform noise filtering and amplitude normalization processing on each current time sequence Seq in the current signal set Cur. The noise filtering is achieved by using a sliding window smoothing method to filter out high-frequency noise for each current time sequence Seq. Subsequently, the current time series Seq after noise filtering is normalized in amplitude. The difference between the maximum and minimum current values ​​in the current time series Seq is used as the amplitude reference, and the current value of each sampling point is linearly mapped to the [0,1] interval to form the normalized current time series SeqN. All normalized current timing sequences SeqN are structurally mapped according to their original pin numbers and period numbers, and are summarized to form a normalized signal set Nor; The normalized signal set Nor and the current signal set Cur maintain a one-to-one correspondence in structure, that is, each current time series Seq has its matching normalized current time series SeqN.

[0009] Preferably, S2 further includes S22; S22. The normalized signal set Nor is used as input. Each normalized current time sequence SeqN is segmented according to time intervals. The dynamic characteristic parameters of the current signal are calculated in each time segment, including: instantaneous rate of change, peak amplitude and period difference. The instantaneous rate of change is obtained by calculating the rate of change of current at continuous sampling points within each time segment, calculating the difference between the current values ​​of two adjacent sampling points, and then dividing by the time interval between them. The peak amplitude is obtained by iterating through all normalized current sample values ​​in each time segment and taking the maximum normalized current value as the peak amplitude. The period difference is calculated by dividing the time into segments at corresponding positions in two adjacent conduction cycles of the same pin, calculating the average current within each segment, and then calculating the difference between the two average values ​​as the period difference value. The instantaneous rate of change, the peak amplitude, and the period difference are arranged in time segment order to form a feature vector for a single pin, which is labeled as the feature set Fea.

[0010] Preferably, S3 includes S31; S31. Using the feature set Fea as the basis for analysis, autocorrelation analysis is performed on the feature vector of each pin. The autocorrelation function value is calculated by taking the product of the current value and the delayed value in the feature vector according to a fixed delay step size. For each delay, the product of the current value and the delayed value in the numerical sequence is taken and averaged to reflect the similarity of the numerical sequence under time shift. By calculating the autocorrelation values ​​within all delay step ranges, the complete autocorrelation function curves of the three dynamic feature parameters in the feature vector are obtained; The eigenvector calculation results of each pin form an autocorrelation curve CorL, and the autocorrelation curves CorL corresponding to all pins form the autocorrelation curve set Cor; Each autocorrelation curve CorL in the set of autocorrelation curves Cor describes the autocorrelation relationship of the pin's eigenvectors under different time delays, and the curve shape can reflect the strength of the signal's periodicity.

[0011] Preferably, S3 further includes S32; S32. Perform characteristic peak identification and delay distribution analysis on each autocorrelation curve CorL in the set of autocorrelation curves Cor; Among them, the characteristic peak identification is performed by performing peak search on each autocorrelation curve CorL to identify the main peak and several secondary peaks; the main peak is located at the zero delay position, corresponding to the autocorrelation energy of the signal itself, and does not participate in the periodic feature analysis; the secondary peaks are distributed in the non-zero delay interval, corresponding to the repetitive correlation of the signal under time shift, which is the basis for periodic disturbance analysis. For secondary peaks, record the delay position Del and peak amplitude Amp; then calculate the delay stability index Dst and amplitude consistency index Ampc between adjacent secondary peaks; The delay stability index Dst is obtained by calculating the standard deviation of the delay position Del interval between adjacent primary secondary peaks; the amplitude consistency index Ampc is obtained by statistically analyzing the peak amplitude Amp of the secondary peaks and calculating the amplitude standard deviation. The delay stability index Dst and the amplitude consistency index Ampc are input into a composite mapping function for calculation to obtain the periodic jitter intensity coefficient Jit.

[0012] Preferably, S4 includes S41; S41. The ratio of the periodic jitter intensity coefficient Jit to the preset standard reference value NorJ is calculated to obtain the deviation ratio Rat; The formula for calculating the deviation ratio Rat is: Rat = (Jit - NorJ) ÷ NorJ.

[0013] Preferably, S4 further includes S42; S42. Compare the deviation ratio Rat with the preset threshold standard Thr to obtain the abnormal judgment of the pin status, and generate the mark of the IGBT package structure based on the abnormal judgment result. The pin status is obtained through the following comparison method: When the deviation ratio Rat ≤ the threshold standard Thr, the pin state is determined to be stable, the conduction cycle disturbance is within the allowable range, and the abnormal judgment result is marked as qualified. When the deviation ratio Rat > the threshold standard Thr, it is determined that there is an abnormal periodic disturbance in the pin, indicating that the IGBT package structure has micro-loosening, solder joint cracks or bonding layer degradation, and the abnormal judgment result is marked as unqualified. During the judgment process, the marking results of all tested pins are summarized to form an anomaly judgment set Abn; the anomaly judgment set Abn contains the number of each pin, the deviation ratio Rat, and the judgment result.

[0014] An intelligent analysis and testing system for IGBT package structure includes an IGBT package pin data acquisition module, a data preprocessing and feature extraction module, a periodic analysis module, and an analysis and testing decision module; The IGBT package pin data acquisition module acquires the dynamic current signal sequence of the IGBT package pin during the on / off test process and constructs a current signal set Cur; The data preprocessing and feature extraction module performs noise filtering, amplitude normalization, and time scale alignment on the current signal set Cur to obtain a normalized signal set Nor. The normalized signal set Nor is divided into segments according to time intervals, and the instantaneous rate of change, peak amplitude, and period difference are calculated for each segment of the waveform to obtain a feature set Fea that reflects the dynamic characteristics of the waveform. The periodic analysis module performs autocorrelation function calculation on the feature set Fea to generate an autocorrelation curve set Cor. By analyzing the peak distribution and delay interval in the autocorrelation curve set Cor, the periodic jitter intensity coefficient Jit is obtained. The analysis and testing decision module compares the periodic jitter intensity coefficient Jit with the preset normal operation model standard value NorJ, calculates the deviation ratio Rat, and determines that the current IGBT package pin is loose when the deviation ratio Rat exceeds the preset judgment threshold Thr, generating an abnormal judgment set Abn.

[0015] This invention provides an intelligent analysis and testing method and system for IGBT packaging structures, which has the following beneficial effects: (1) By constructing a current signal set Cur and extracting its normalized features to form a feature set Fea, and then generating an autocorrelation curve set Cor through autocorrelation operation, the periodic jitter intensity coefficient Jit is calculated, realizing the cross-domain mapping from electrical signal changes to mechanical loosening state. Unlike the traditional method that relies on external vibration loading or manual judgment, this method can identify early loosening signs of pins through small changes in the periodic jitter intensity coefficient Jit under normal electrical characteristic test conditions. Furthermore, by comparing with the normal jitter standard value NorJ of the normal model, the deviation ratio Rat is calculated, and an abnormal judgment set Abn is formed according to the set judgment threshold Thr, thereby realizing automated abnormal identification and result classification. It not only avoids the problems of large external excitation equipment, long detection delay, and insufficient sensitivity in traditional detection methods, but also enables online detection and health warning when the IGBT packaged module is still in operation. It realizes early identification of potential failures and maintenance decision support.

[0016] (2) By calculating the autocorrelation function of each pin's eigenvector based on the feature set Fea and forming an autocorrelation curve set Cor, the repeatability and consistency characteristics of the signal under time shift can be revealed. Furthermore, by performing peak identification and delay distribution analysis on the secondary peaks in the autocorrelation curve set Cor, the delay stability index Dst and amplitude consistency index Ampc are calculated, and then the periodic jitter intensity coefficient Jit is obtained, thereby realizing a quantitative expression of the pin conduction characteristic stability. This method can transform the extremely subtle repetitive disturbances in the electrical signal changes into measurable parameters, breaking through the limitations of previous reliance on spectrum analysis or manual waveform comparison. It realizes the electrical signal response to mechanical loosening without external excitation, and through the fusion analysis of multi-dimensional correlation features, it makes early diagnosis of hidden structural fatigue possible, greatly improving the reliability monitoring sensitivity and predictive analysis capability of IGBT package structure during operation.

[0017] (3) By comparing the periodic jitter intensity coefficient Jit with the established normal jitter standard value NorJ, the deviation ratio Rat is calculated, which can quantitatively reflect the degradation amplitude of pin signal stability. Then, by comparing it with the judgment threshold Thr, an abnormal judgment set Abn is generated, realizing the mapping from continuous feature quantity to discrete health state. This judgment logic has adaptive recognition capability and can automatically distinguish between normal disturbance and abnormal looseness in complex environments. This early recognition mechanism transforms package reliability from passive detection to active early warning. The detection results not only output qualified and unqualified status indicators, but also intuitively reflect the hierarchical differences in the degradation degree of each pin, thereby providing accurate decision-making basis for equipment operation and maintenance. It realizes the full-link closed-loop control of IGBT package structure from testing and measurement to intelligent judgment, making the package health assessment have the characteristics of being quantifiable, traceable and predictable. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the steps of an intelligent analysis and testing method for an IGBT packaging structure according to the present invention; Figure 2 This is a schematic block diagram of an intelligent analysis and testing system for an IGBT packaging structure according to the present invention. Figure 3 This is a schematic diagram showing the trend relationship between the periodic jitter intensity coefficient Jit and the normal jitter standard value NorJ. Detailed Implementation

[0019] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention. Example

[0020] This invention provides an intelligent analysis and testing method for IGBT package structures. Please refer to [link / reference]. Figure 1 This includes the following steps: S1. Collect the dynamic current signal sequence of the IGBT package pins during the on / off test process and construct the current signal set Cur; S2. Perform noise filtering, amplitude normalization, and time scale alignment on the current signal set Cur to obtain the normalized signal set Nor. Divide the normalized signal set Nor into segments according to the time interval, calculate the instantaneous rate of change, peak amplitude, and period difference for each segment of the waveform, and obtain the feature set Fea used to reflect the dynamic characteristics of the waveform. S3. Perform autocorrelation function calculation on feature set Fea to generate autocorrelation curve set Cor. By analyzing the peak distribution and delay interval in autocorrelation curve set Cor, obtain the periodic jitter intensity coefficient Jit. S4. Compare the periodic jitter intensity coefficient Jit with the preset normal operation model standard value NorJ, calculate the deviation ratio Rat, and when the deviation ratio Rat exceeds the preset judgment threshold Thr, determine that the current IGBT package pin is in a loose state and generate an abnormal judgment set Abn.

[0021] In this embodiment, a current signal set Cur is constructed and its normalized features are extracted to form a feature set Fea. Then, an autocorrelation curve set Cor is generated through autocorrelation calculation, and the periodic jitter intensity coefficient Jit is calculated accordingly, achieving a cross-domain mapping from electrical signal changes to mechanical loosening states. Unlike traditional methods that rely on external vibration loading or manual judgment, this method can identify early signs of pin loosening through minute changes in the periodic jitter intensity coefficient Jit under normal electrical characteristic testing conditions. Furthermore, by comparing with the normal jitter standard value NorJ of a normal model, the deviation ratio Rat is calculated, and an anomaly judgment set Abn is formed based on a set judgment threshold Thr, thereby achieving automated anomaly identification and result classification. This method not only avoids the problems of large external excitation equipment, long detection delay, and insufficient sensitivity in traditional detection methods, but also enables online detection and health warning while the IGBT packaged module is still in operation. For example, in the long-term vibration environment of industrial frequency converter modules, when micron-level structural loosening occurs at the pin solder joints, the periodic jitter intensity coefficient Jit will show a slight increase. Based on this, it can be determined that the deviation ratio Rat exceeds the judgment threshold Thr and is automatically marked as "unqualified" in the anomaly judgment set Abn, enabling early identification of potential failures and support for maintenance decisions. Therefore, this invention transforms pin contact reliability testing from static offline testing to dynamic intelligent judgment, significantly improving the safety, predictability, and maintenance efficiency of IGBT package structures in complex application scenarios. Example

[0022] Specifically: S1 includes S11; S11. During the IGBT package structure continuity test, a sampling circuit for measuring current is connected in series on the pin power path to sample the current during the continuity process of each pin. During the current sampling process, the current is sampled multiple times at fixed time intervals. The time interval is set to between one-tenth and one-twentieth of the IGBT switching frequency to ensure that the sampling process covers the entire on and off cycle. During sampling, the sampling time, instantaneous current value, corresponding pin number and conduction cycle number are recorded respectively. The continuous sampling data are arranged in chronological order to form the raw current sampling set Raw. When a sudden change in the instantaneous current value is detected between adjacent sampling points, a moving average smoothing is performed on the sudden change segment to avoid error amplification caused by electromagnetic interference. The raw current sampling set includes continuous current variation data for each pin over multiple switching cycles, which is used to reflect the actual current response of the pin under different conduction states; It should be noted that: Sampling is completed directly through the current sampling loop in the pin path. The current sampling loop can be composed of current sensing elements (such as shunt resistors or Hall elements). The signal is output in the form of voltage and then the value is acquired. Sampling frequency definition: The sampling frequency is more than ten times the switching frequency, so that the sampling sequence can completely cover the dynamic process of the turn-on and turn-off phases in the time domain.

[0023] S1 further includes S12; S12. Group the original current sampling set Raw according to the pin number and the cycle number, sort the sampling points of each group in time order, so that the current change in each cycle is continuous without interruption, and perform time axis interpolation on the sorted data at equal intervals to correct the sampling interval error and obtain a smooth current time sequence Seq. Then, the current timing sequence Seq of the same pin under multiple cycles is uniformly organized, and all time sequences are spliced ​​together to form a complete pin current response sequence for each pin. Then, the current timing sequences Seq corresponding to all pins are summarized to generate the current signal set Cur. It should be noted that: The current signal set Cur consists of multiple current time series Seq. Each current time series Seq represents the current response of a single pin under multiple switching cycles. The data format maintains time continuity and amplitude consistency, which facilitates direct feature extraction and autocorrelation analysis. The sampling time interval is corrected by using time sorting and linear interpolation to ensure that all sampling points are equidistant in the time domain; High-frequency noise is eliminated by using the moving average method, so that the current time series Seq remains smooth and continuous; Through time-series reconstruction and interpolation correction, the obtained current time-series sequence Seq maintains consistency in both time and amplitude, with signal distortion of less than 1%. The current signal set Cur ensures that parameters such as the rate of change and waveform deviation calculated in subsequent feature extraction have a unified reference benchmark, thereby improving the analytical accuracy of the detection method.

[0024] In this embodiment, a raw current sampling set (Raw) that can be directly quantified is formed by connecting a current sampling loop in series in the pin energizing path. Subsequently, a structured current timing sequence (Seq) is generated through time interpolation and smoothing, and then summarized into a current signal set (Cur). This multi-layer data reconstruction method ensures the continuity and comparability of the sampled signals in both the time and amplitude domains, significantly reducing the impact of electromagnetic noise and sampling interval errors. Unlike previous methods that relied on fixed-point detection or discrete waveform comparison, this scheme can completely capture the dynamic response of the entire IGBT pin conduction process under high-frequency switching conditions. For example, when a power module operates under high-speed pulse drive, even if there is a nanosecond-level transient contact failure at the pin solder joint, the abnormal abrupt change segment can still be identified through the smooth change of the current timing sequence (Seq), thus reflecting potential structural instability at the signal level. Therefore, this invention upgrades IGBT packaging testing from "single-point sampling" to "full-cycle dynamic monitoring," which not only improves the temporal resolution and amplitude accuracy of the test results, but also provides a high-quality input foundation for subsequent feature set Fea extraction and autocorrelation curve set Cor analysis. It is particularly suitable for real-time health monitoring and packaging reliability analysis under conditions such as high-speed switching and high-frequency vibration. Example

[0025] Specifically: S2 includes S21; S21. Perform noise filtering and amplitude normalization processing on each current time sequence Seq in the current signal set Cur. The noise filtering is achieved by using a sliding window smoothing method to filter out high-frequency noise for each current time sequence Seq. The window width is determined according to the number of sampling points, and is generally taken as 5% to 10% of the number of sampling points in a single cycle, in order to preserve the main waveform and suppress random jitter. Subsequently, the current time series Seq after noise filtering is normalized in amplitude. The difference between the maximum and minimum current values ​​in the current time series Seq is used as the amplitude reference, and the current value of each sampling point is linearly mapped to the [0,1] interval to form the normalized current time series SeqN. All normalized current timing sequences SeqN are structurally mapped according to their original pin numbers and period numbers, and are summarized to form a normalized signal set Nor; The normalized signal set Nor and the current signal set Cur maintain a one-to-one correspondence in structure, that is, each current time series Seq has its matching normalized current time series SeqN. It should be noted that: Noise removal employs a moving average filtering method for smoothing, calculated by taking the average of all sampling points within the preceding and following windows at each time point as the output for the current point; this effectively suppresses sudden noise and maintains the waveform trend. Normalization calculation method: The normalized current value at each sampling point is obtained through the following relationship: Normalized value = (Sampling point current - Minimum current) / (Maximum current - Minimum current); Normalization eliminates amplitude differences between different channels, allowing waveform feature analysis to reflect only dynamic changes without being affected by absolute amplitude.

[0026] S2 further includes S22; S22. The normalized signal set Nor is used as input. Each normalized current time sequence SeqN is segmented according to time intervals. The dynamic characteristic parameters of the current signal are calculated in each time segment, including: instantaneous rate of change, peak amplitude and period difference. The instantaneous rate of change is calculated by taking continuous sampling points within each time segment to determine the rate of change of current. The difference between the current values ​​of two adjacent sampling points is calculated and then divided by the time interval between them to obtain the instantaneous rate of change. It is used to reflect the speed at which the current rises or falls. The larger the value, the more drastic the current change, indicating that there is a fluctuation in the pin's conduction characteristics. The peak amplitude is obtained by iterating through all normalized current sample values ​​in each time segment and taking the maximum normalized current value as the peak amplitude; it is used to represent the maximum stable amplitude of the current response during conduction. If the peak amplitude of a certain segment is significantly lower than the normal range, it indicates that there is an increase in the conduction contact impedance or a discontinuity in conduction at the pin. The period difference is calculated by dividing the time into segments at corresponding positions in two adjacent conduction cycles of the same pin, and calculating the average current in each segment; then the difference between the two average values ​​is calculated as the period difference value; it is used to reflect the consistency change of the pin in different conduction cycles. The larger the period difference, the higher the instability of the pin conduction characteristics, which may be due to contact fluctuations caused by mechanical loosening or thermal fatigue. The instantaneous rate of change, the peak amplitude, and the period difference are arranged in time segment order to form a feature vector for a single pin, which is labeled as the feature set Fea.

[0027] In this embodiment, noise filtering and amplitude normalization are performed on each current time sequence Seq in the current signal set Cur to obtain a structure-matched normalized signal set Nor. This ensures the consistency of the signals of different pin channels in terms of amplitude and time scale, allowing subsequent analysis to be based entirely on relative change characteristics rather than absolute amplitude differences. Simultaneously, by segmenting the normalized signal set Nor by time and calculating the instantaneous rate of change, peak amplitude, and period difference, a feature set Fea is formed, enabling a quantitative expression of the signal dynamic response across multiple dimensions. This processing not only eliminates the influence of gain differences and sensor offsets between test devices but also significantly improves the characteristic sensitivity of periodic waveforms. For example, when an IGBT module operates in a high-temperature environment, causing a slight change in pin on-resistance, traditional amplitude monitoring methods struggle to distinguish between thermal noise and mechanical drift. However, this method can accurately reflect this decrease in conduction stability through abnormal fluctuations in the instantaneous rate of change and slight shifts in the period difference. Thus, while maintaining the integrity of the waveform morphology, step S2 achieves the standardization and high-resolution extraction of signal features, making the subsequent autocorrelation curve set Cor analysis more stable and comparable, thereby providing key data support for identifying latent degradations such as poor pin contact and conduction fluctuations. Example

[0028] Specifically: S3 includes S31; S31. Using the feature set Fea as the basis for analysis, autocorrelation analysis is performed on the feature vector of each pin. The autocorrelation function value is calculated by taking the product of the current value and the delayed value in the feature vector according to a fixed delay step size. For each delay, the product of the current value and the delayed value in the numerical sequence is taken and averaged to reflect the similarity of the numerical sequence under time shift. By calculating the autocorrelation values ​​across all delay step ranges (from zero delay to single-cycle length), the complete autocorrelation function curves of the three dynamic characteristic parameters in the eigenvector are obtained; The eigenvector calculation results of each pin form an autocorrelation curve CorL, and the autocorrelation curves CorL corresponding to all pins form the autocorrelation curve set Cor; Each autocorrelation curve CorL in the set of autocorrelation curves Cor describes the autocorrelation relationship of the pin's eigenvectors under different time delays, and the curve shape can reflect the strength of the signal's periodicity. It should be noted that: Definition and principle of autocorrelation function: The autocorrelation function is a statistical function that measures the similarity between a signal and its time-shifted version. It is used to reveal the repeatability and correlation of a signal in the time domain. For any time series signal, if it still shows high correlation after a certain time interval, it indicates that there is a periodic component in the signal. In this method, the autocorrelation analysis of the eigenvectors of each pin can capture the repetitive patterns of the pin current waveform during multi-cycle conduction, thereby revealing whether its mechanical or electrical characteristics are stable. When the pin connection is stable and the conduction characteristics remain consistent throughout each cycle, the corresponding autocorrelation curve CorL shows a smooth decreasing trend at multiple delay points, and there are almost no obvious secondary peaks except for zero delay. When there is slight mechanical looseness or solder joint fatigue in the pin, the current characteristic sequence of each cycle will produce a subtle periodic perturbation. This perturbation will appear as a distinct secondary peak in the autocorrelation function, with its delay position corresponding to the cycle length and its amplitude reflecting the perturbation strength. Therefore, the autocorrelation curve CorL can not only reveal the repeatability of the signal, but also quantify the strength of its perturbation. This method is particularly suitable for detecting low-amplitude, non-fixed-period mechanical fluctuations generated by IGBT pins during long-term operation; its advantage is that it can quantitatively identify mechanical loosening through the statistical characteristics of the signal itself without the need for additional excitation or vibration equipment.

[0029] S3 further includes S32; S32. Perform characteristic peak identification and delay distribution analysis on each autocorrelation curve CorL in the set of autocorrelation curves Cor; Among them, the characteristic peak identification is performed by performing peak search on each autocorrelation curve CorL to identify the main peak and several secondary peaks; the main peak is located at the zero delay position, corresponding to the autocorrelation energy of the signal itself, and does not participate in the periodic feature analysis; the secondary peaks are distributed in the non-zero delay interval, corresponding to the repetitive correlation of the signal under time shift, which is the basis for periodic disturbance analysis. Among the identified peaks, secondary peaks are selected as the objects of analysis, and the selection principle is that the following conditions must be met simultaneously: Condition 1: The peak amplitude (Amp) is more than 30% higher than the average amplitude of the entire curve; Condition 2: The corresponding delay position Del is within a range of multiple cycles after zero delay; The delay position Del of the secondary peaks that satisfy conditions one and two, together with the peak amplitude Amp, form a peak distribution dataset, which is used as the analysis object. For secondary peaks, record the delay position Del and peak amplitude Amp; then calculate the delay stability index Dst and amplitude consistency index Ampc between adjacent secondary peaks; The delay stability index Dst is obtained by calculating the standard deviation of the interval between the delay positions Del of adjacent major secondary peaks. The smaller the standard deviation, the more stable the repetition during the period; conversely, the larger the standard deviation, the more unstable the periodic fluctuation of the signal. The delay stability index Dst is used to characterize the repetitive consistency of the pin signal on the time scale. The specific calculation process is as follows: calculate the delay interval value between each two adjacent secondary peaks, then calculate the average value of all delay interval values, and then use this average value as a reference to calculate the deviation of each delay interval value from the average value. After squaring these deviations, calculate the average value, and then take the square root. The result is the delay stability index Dst. The amplitude consistency index Ampc statistically analyzes the peak amplitude Amp of the secondary peaks and calculates the amplitude standard deviation. The smaller the amplitude variation, the better the energy consistency of the periodic waveform; the larger the variation, the more energy drift or amplitude attenuation occurs in the conduction waveform. The amplitude consistency index Ampc is used to describe the consistency of signal energy between periods. The specific calculation process is as follows: calculate the average peak amplitude of all major secondary peaks, then calculate the deviation of the peak amplitude of each secondary peak from the average value, square these deviations and average them, then take the square root to obtain the amplitude consistency index Ampc. The delay stability index Dst and the amplitude consistency index Ampc are input into a composite mapping function for calculation to obtain the periodic jitter intensity coefficient Jit. The specific calculation formula is as follows: The periodic jitter intensity coefficient Jit = α × [Dst ÷ (Dst + k1)] + β × [Ampc ÷ (Ampc + k2)]; In the formula, α and β represent the weighting coefficients of the delay stability index Dst and the amplitude consistency index Ampc, respectively, and the specific values ​​are set by the user, and α+β; k1 and k2 represent the scale correction coefficients of the delay stability index Dst and the amplitude consistency index Ampc, respectively, used to prevent calculation bias caused by minimum values; the formal meaning of this formula is that when either the delay stability index Dst or the amplitude consistency index Ampc increases, the periodic jitter intensity coefficient Jit will increase synchronously. When both are close to zero, the periodic jitter intensity coefficient Jit approaches zero, indicating that the signal period is stable and there is no obvious mechanical disturbance.

[0030] In this embodiment, by calculating the autocorrelation function of each pin's eigenvector based on the feature set Fea and forming an autocorrelation curve set Cor, the repeatability and consistency characteristics of the signal under time shift can be revealed. Furthermore, by performing peak identification and delay distribution analysis on the secondary peaks in the autocorrelation curve set Cor, the delay stability index Dst and the amplitude consistency index Ampc are calculated, and then the periodic jitter intensity coefficient Jit is obtained, thereby achieving a quantitative expression of the pin's conduction characteristic stability. This method can transform extremely subtle repetitive disturbances in electrical signal changes into measurable parameters, overcoming the limitations of previous methods that relied on spectrum analysis or manual waveform comparison. For example, in the long-term operating environment of high-speed railway traction inverters, IGBT module pins may experience micro-level gap loosening due to thermal cycling and mechanical vibration, which cannot be detected by traditional DC resistance detection. However, this method can reflect this metastable disturbance trend through a slight increase in the periodic jitter intensity coefficient Jit. When the delay stability index Dst and the amplitude consistency index Ampc increase simultaneously, the periodic drift and energy decay characteristics of the conduction waveform can be accurately identified, thereby determining that the pin's mechanical characteristics have shown signs of instability. Therefore, this step not only achieves electrical signal response to mechanical loosening without external excitation, but also enables early diagnosis of hidden structural fatigue through the fusion analysis of multi-dimensional related features, greatly improving the reliability monitoring sensitivity and predictive analysis capability of IGBT package structure during operation. Example

[0031] Please see Figure 1 and Figure 3 Specifically: S4 includes S41; S41. The ratio of the periodic jitter intensity coefficient Jit to the preset standard reference value NorJ is calculated to obtain the deviation ratio Rat. The deviation ratio Rat is the ratio of the actual measured periodic disturbance degree to the normal model reference level, which is used to represent the degradation magnitude of the pin signal stability. The formula for calculating the deviation ratio Rat is: Rat = (Jit - NorJ) ÷ NorJ; It should be noted that: When the IGBT package structure is intact and the pin contact is stable, the measured periodic jitter intensity coefficient Jit is close to the standard reference value NorJ, and the corresponding deviation ratio Rat is close to zero.

[0032] S4 also includes S42; S42. Compare the deviation ratio Rat with the preset threshold standard Thr to obtain the abnormal judgment of the pin status, and generate the mark of the IGBT package structure based on the abnormal judgment result. The pin status is obtained through the following comparison method: When the deviation ratio Rat ≤ the threshold standard Thr, the pin state is determined to be stable, the conduction cycle disturbance is within the allowable range, and the abnormal judgment result is marked as qualified. When the deviation ratio Rat > the threshold standard Thr, it is determined that there is an abnormal periodic disturbance in the pin, indicating that the IGBT package structure has micro-loosening, solder joint cracks or bonding layer degradation, and the abnormal judgment result is marked as unqualified. During the judgment process, the marking results of all tested pins are summarized to form an anomaly judgment set Abn; the anomaly judgment set Abn contains the number of each pin, the deviation ratio Rat, and the judgment result.

[0033] In this embodiment, the deviation ratio Rat is calculated by comparing the periodic jitter intensity coefficient Jit with the established normal jitter standard value NorJ. This method can quantitatively reflect the degradation range of pin signal stability. Then, by comparing it with the judgment threshold Thr, an anomaly judgment set Abn is generated, realizing the mapping from continuous characteristic quantities to discrete health states. This judgment logic has adaptive recognition capabilities, automatically distinguishing between normal disturbances and abnormal loosening in complex environments. For example, when the IGBT package structure in a wind power converter experiences slight solder joint fatigue due to long-term mechanical vibration, traditional detection can only show current anomalies after significant failure. However, this method can determine a potential risk and mark the corresponding pin as "unqualified" in the anomaly judgment set Abn when the deviation ratio Rat is slightly higher than the judgment threshold Thr. This early identification mechanism shifts package reliability from passive detection to proactive early warning. The detection results not only output qualified and unqualified status indicators but also intuitively reflect the hierarchical differences in the degradation degree of each pin, thus providing accurate decision-making basis for equipment operation and maintenance. It realizes closed-loop control of the entire IGBT packaging structure from testing and measurement to intelligent judgment, making the packaging health assessment quantifiable, traceable and predictable. Example

[0034] A smart analysis and testing system for IGBT packaging structures, please refer to... Figure 2 Specifically, it includes an IGBT package pin data acquisition module, a data preprocessing and feature extraction module, a periodic analysis module, and an analysis and testing decision module; The IGBT package pin data acquisition module acquires the dynamic current signal sequence of the IGBT package pin during the on / off test process and constructs a current signal set Cur; The data preprocessing and feature extraction module performs noise filtering, amplitude normalization, and time scale alignment on the current signal set Cur to obtain a normalized signal set Nor. The normalized signal set Nor is divided into segments according to time intervals, and the instantaneous rate of change, peak amplitude, and period difference are calculated for each segment of the waveform to obtain a feature set Fea that reflects the dynamic characteristics of the waveform. The periodic analysis module performs autocorrelation function calculation on the feature set Fea to generate an autocorrelation curve set Cor. By analyzing the peak distribution and delay interval in the autocorrelation curve set Cor, the periodic jitter intensity coefficient Jit is obtained. The analysis and testing decision module compares the periodic jitter intensity coefficient Jit with the preset normal operation model standard value NorJ, calculates the deviation ratio Rat, and determines that the current IGBT package pin is loose when the deviation ratio Rat exceeds the preset judgment threshold Thr, generating an abnormal judgment set Abn.

[0035] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A smart analysis and testing method for IGBT packaging structure, characterized in that: Includes the following steps: S1. Collect the dynamic current signal sequence of the IGBT package pins during the on / off test process and construct the current signal set Cur; S2. Perform noise filtering, amplitude normalization, and time scale alignment on the current signal set Cur to obtain the normalized signal set Nor. Divide the normalized signal set Nor into segments according to the time interval, calculate the instantaneous rate of change, peak amplitude, and period difference for each segment of the waveform, and obtain the feature set Fea used to reflect the dynamic characteristics of the waveform. S3. Perform autocorrelation function calculation on feature set Fea to generate autocorrelation curve set Cor. By analyzing the peak distribution and delay interval in autocorrelation curve set Cor, obtain the periodic jitter intensity coefficient Jit. S4. Compare the periodic jitter intensity coefficient Jit with the preset normal operation model standard value NorJ, calculate the deviation ratio Rat, and when the deviation ratio Rat exceeds the preset judgment threshold Thr, determine that the current IGBT package pin is in a loose state and generate an abnormal judgment set Abn.

2. The intelligent analysis and testing method for IGBT packaging structure according to claim 1, characterized in that: S1 includes S11; S11. During the IGBT package structure continuity test, a sampling circuit for measuring current is connected in series on the pin power path to sample the current during the continuity process of each pin. During the current sampling process, the current is sampled multiple times at fixed time intervals. The time interval is set to between one-tenth and one-twentieth of the IGBT switching frequency. During sampling, the sampling time, instantaneous current value, corresponding pin number and conduction cycle number are recorded respectively. The continuous sampling data are arranged in chronological order to form the raw current sampling set Raw. The Raw current sampling set includes continuous current variation data for each pin over multiple switching cycles.

3. The intelligent analysis and testing method for IGBT packaging structure according to claim 2, characterized in that: S1 further includes S12; S12. Group the original current sampling set Raw according to the pin number and the cycle number, sort the sampling points of each group in time order, so that the current change in each cycle is continuous without interruption, and perform time axis interpolation on the sorted data at equal intervals to correct the sampling interval error and obtain a smooth current time sequence Seq. Then, the current timing sequence Seq of the same pin under multiple cycles is uniformly organized, and all time sequences are spliced ​​together to form a complete pin current response sequence for each pin. Then, the current timing sequences Seq corresponding to all pins are summarized to generate the current signal set Cur.

4. The intelligent analysis and testing method for IGBT packaging structure according to claim 3, characterized in that: S2 includes S21; S21. Perform noise filtering and amplitude normalization processing on each current time sequence Seq in the current signal set Cur. The noise filtering is achieved by using a sliding window smoothing method to filter out high-frequency noise for each current time sequence Seq. Subsequently, the current time series Seq after noise filtering is normalized in amplitude. The difference between the maximum and minimum current values ​​in the current time series Seq is used as the amplitude reference, and the current value of each sampling point is linearly mapped to the [0,1] interval to form the normalized current time series SeqN. All normalized current timing sequences SeqN are structurally mapped according to their original pin numbers and period numbers, and are summarized to form a normalized signal set Nor; The normalized signal set Nor and the current signal set Cur maintain a one-to-one correspondence in structure, that is, each current time series Seq has its matching normalized current time series SeqN.

5. The intelligent analysis and testing method for IGBT packaging structure according to claim 4, characterized in that: S2 further includes S22; S22. The normalized signal set Nor is used as input. Each normalized current time sequence SeqN is segmented according to time intervals. The dynamic characteristic parameters of the current signal are calculated in each time segment, including: instantaneous rate of change, peak amplitude and period difference. The instantaneous rate of change is obtained by calculating the rate of change of current at continuous sampling points within each time segment, calculating the difference between the current values ​​of two adjacent sampling points, and then dividing by the time interval between them. The peak amplitude is obtained by iterating through all normalized current sample values ​​in each time segment and taking the maximum normalized current value as the peak amplitude. The period difference is calculated by dividing the time into segments at corresponding positions in two adjacent conduction cycles of the same pin, calculating the average current within each segment, and then calculating the difference between the two average values ​​as the period difference value. The instantaneous rate of change, the peak amplitude, and the period difference are arranged in time segment order to form a feature vector for a single pin, which is labeled as the feature set Fea.

6. The intelligent analysis and testing method for IGBT packaging structure according to claim 5, characterized in that: S3 includes S31; S31. Using the feature set Fea as the basis for analysis, autocorrelation analysis is performed on the feature vector of each pin. The autocorrelation function value is calculated by taking the product of the current value and the delayed value in the feature vector according to a fixed delay step size. For each delay, the product of the current value and the delayed value in the numerical sequence is taken and averaged to reflect the similarity of the numerical sequence under time shift. By calculating the autocorrelation values ​​within all delay step ranges, the complete autocorrelation function curves of the three dynamic feature parameters in the feature vector are obtained; The eigenvector calculation results of each pin form an autocorrelation curve CorL, and the autocorrelation curves CorL corresponding to all pins form the autocorrelation curve set Cor; Each autocorrelation curve CorL in the set of autocorrelation curves Cor describes the autocorrelation relationship of the pin's eigenvectors under different time delays, and the curve shape can reflect the strength of the signal's periodicity.

7. The intelligent analysis and testing method for IGBT packaging structure according to claim 6, characterized in that: S3 further includes S32; S32. Perform characteristic peak identification and delay distribution analysis on each autocorrelation curve CorL in the set of autocorrelation curves Cor; Among them, the characteristic peak identification is performed by performing peak search on each autocorrelation curve CorL to identify the main peak and several secondary peaks; the main peak is located at the zero delay position, corresponding to the autocorrelation energy of the signal itself, and does not participate in the periodic feature analysis; the secondary peaks are distributed in the non-zero delay interval, corresponding to the repetitive correlation of the signal under time shift, which is the basis for periodic disturbance analysis. For secondary peaks, record the delay position Del and peak amplitude Amp; then calculate the delay stability index Dst and amplitude consistency index Ampc between adjacent secondary peaks; The delay stability index Dst is obtained by calculating the standard deviation of the delay position Del interval between adjacent major secondary peaks; The amplitude consistency index Ampc statistically analyzes the peak amplitude Amp of the secondary peak and calculates the amplitude standard deviation. The delay stability index Dst and the amplitude consistency index Ampc are input into a composite mapping function for calculation to obtain the periodic jitter intensity coefficient Jit.

8. The intelligent analysis and testing method for IGBT packaging structure according to claim 7, characterized in that: S4 includes S41; S41. The ratio of the periodic jitter intensity coefficient Jit to the preset standard reference value NorJ is calculated to obtain the deviation ratio Rat; The formula for calculating the deviation ratio Rat is: Rat = (Jit - NorJ) ÷ NorJ.

9. The intelligent analysis and testing method for IGBT packaging structure according to claim 8, characterized in that: S4 also includes S42; S42. Compare the deviation ratio Rat with the preset threshold standard Thr to obtain the abnormal judgment of the pin status, and generate the mark of the IGBT package structure based on the abnormal judgment result. The pin status is obtained through the following comparison method: When the deviation ratio Rat ≤ the threshold standard Thr, the pin state is determined to be stable, the conduction cycle disturbance is within the allowable range, and the abnormal judgment result is marked as qualified. When the deviation ratio Rat > the threshold standard Thr, it is determined that there is an abnormal periodic disturbance in the pin, indicating that the IGBT package structure has micro-loosening, solder joint cracks or bonding layer degradation, and the abnormal judgment result is marked as unqualified. During the judgment process, the marking results of all tested pins are summarized to form an anomaly judgment set Abn; the anomaly judgment set Abn contains the number of each pin, the deviation ratio Rat, and the judgment result.

10. An intelligent analysis and testing system for IGBT package structure, applied to the intelligent analysis and testing method for IGBT package structure according to any one of claims 1 to 9, characterized in that: It includes an IGBT package pin data acquisition module, a data preprocessing and feature extraction module, a periodic analysis module, and an analysis and testing decision module; The IGBT package pin data acquisition module acquires the dynamic current signal sequence of the IGBT package pin during the on / off test process and constructs a current signal set Cur; The data preprocessing and feature extraction module performs noise filtering, amplitude normalization, and time scale alignment on the current signal set Cur to obtain a normalized signal set Nor. The normalized signal set Nor is divided into segments according to time intervals, and the instantaneous rate of change, peak amplitude, and period difference are calculated for each segment of the waveform to obtain a feature set Fea that reflects the dynamic characteristics of the waveform. The periodic analysis module performs autocorrelation function calculation on the feature set Fea to generate an autocorrelation curve set Cor. By analyzing the peak distribution and delay interval in the autocorrelation curve set Cor, the periodic jitter intensity coefficient Jit is obtained. The analysis and testing decision module compares the periodic jitter intensity coefficient Jit with the preset normal operation model standard value NorJ, calculates the deviation ratio Rat, and determines that the current IGBT package pin is loose when the deviation ratio Rat exceeds the preset judgment threshold Thr, generating an abnormal judgment set Abn.

Images