Diode package defect traceability method and system based on correlation analysis

Abstract

The application provides a diode packaging defect tracing method and system based on correlation analysis, relates to the technical field of electronic packaging process optimization and quality control, and establishes a correlation model for process parameters and X-ray imaging images in the diode packaging process, extracts frequency domain characteristic values of the X-ray images, and performs correlation analysis combined with the process parameters to determine the key abnormal process parameter combination causing the cavity defect. In this way, the accuracy and reliability of defect tracing can be comprehensively improved, which helps to quickly locate abnormal process parameters, optimize packaging process conditions, thereby effectively reducing the occurrence rate of cavity defects, improving the quality and performance of diode packaging, and to a certain extent, solving the problem that the root process parameters of the cavity defect caused by the mismatch between the welding temperature and the pressure are difficult to accurately locate.

Description

Technical Field

[0001] This invention relates to the field of electronic packaging process optimization and quality control technology, and more specifically, to a method and system for tracing the source of diode packaging defects based on correlation analysis. Background Technology

[0002] With the miniaturization and high integration of electronic devices, the quality requirements for diode packaging technology are increasingly stringent. During the packaging process, void defects are one of the key issues affecting diode performance and reliability. The presence of voids leads to increased thermal resistance, decreased electrical performance, and even device failure. Currently, the detection and tracing technology for diode packaging defects mainly relies on X-ray imaging inspection methods, identifying void defect characteristics through imaging analysis of the soldering area. These methods primarily focus on extracting direct parameters such as void area and distribution characteristics from the imaging results, neglecting the deep correlation between void defects and packaging process parameters. Furthermore, some research has begun to attempt to combine machine learning methods to model the process parameters of the packaging process to predict process conditions that may lead to defects. However, in practical applications, these methods often fail to accurately locate the root cause of defects due to a lack of precise characterization of the complex interactions between multiple parameters. Therefore, how to effectively combine packaging process parameters and defect detection data to establish a comprehensive defect tracing method has become an important research direction in the field of diode packaging.

[0003] Existing diode packaging defect tracing technologies have the following shortcomings. First, in defect detection, traditional methods mostly only perform simple analysis of geometric features in X-ray images, without fully exploring the frequency domain information of the images. This results in an inability to comprehensively characterize the characteristics of void defects. Second, in the correlation analysis between process parameters and defects, existing methods typically employ a single linear correlation analysis, which is insufficient to reveal the multi-dimensional impact of complex process parameters on defect formation. Furthermore, existing tracing systems lack effective mechanisms for identifying abnormal parameters, often relying solely on manual experience or simple threshold settings, which can easily lead to insufficient accuracy and reliability of tracing results. In actual packaging processes, key process factors such as soldering temperature and pressure parameters typically exhibit significant nonlinear and dynamic characteristics. However, existing technologies, when constructing process parameter distribution models, fail to fully consider the historical distribution of these parameters and their boundary characteristics under abnormal conditions, resulting in an inability to accurately identify abnormal combinations of process parameters and their contribution to void defects. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention is proposed. This invention provides a method and system for tracing the source of diode packaging defects based on correlation analysis, which to some extent solves the problem of accurately locating the root cause process parameters of void defects caused by mismatched soldering temperature and pressure.

[0005] According to one aspect of the present invention, a method for tracing the source of diode package defects based on correlation analysis is provided, comprising: Collect process parameters during diode packaging and corresponding X-ray imaging images, and establish a database of the correspondence between the process parameters and X-ray imaging images; The X-ray imaging image is subjected to Fourier transform processing to extract the frequency domain feature values ​​of the image. The high-frequency and low-frequency components of the frequency domain feature values ​​are obtained by wavelet decomposition. The ratio of the high-frequency component to the low-frequency component is used as the void defect feature index. Based on the process parameters and the void defect characteristic index, the contribution of welding temperature parameters and pressure parameters to void formation is calculated. Process parameters with correlation coefficients greater than a preset threshold are identified as key influencing factors. K-means clustering boundary values ​​are constructed based on the historical distribution of the key influencing factors. When a new diode package void defect is detected, its void defect characteristic index is extracted, and the corresponding welding temperature parameters and pressure parameters are matched with the cluster boundary value to determine the abnormal process parameter combination that causes the void defect.

[0006] Further, calculating the void defect characteristic index includes: Calculate the frequency domain eigenvalue matrix after Fourier transform. And perform weighted processing; The weighted frequency domain features are subjected to three-level Haar wavelet decomposition to obtain the corresponding high-frequency sub-band and low-frequency sub-band. Energy calculations are performed based on the high-frequency subband and the low-frequency subband respectively to obtain the high-frequency component value and the low-frequency component value; The ratio of the high-frequency component value to the low-frequency component value is calculated, and the dynamic change information is integrated to obtain the void defect characteristic index.

[0007] Furthermore, the frequency domain eigenvalue matrix The calculation is shown in the following formula: in, and These represent the number of rows and columns of the image, respectively. and For frequency domain coordinates, Spatial domain coordinates, It is the imaginary unit.

[0008] Furthermore, the dynamic change information is the rate of change of high-frequency and low-frequency components in continuously acquired images, and the ratio of this rate to the high-frequency and low-frequency component values ​​is integrated with a dynamic adjustment factor to obtain the void defect characteristic index.

[0009] Furthermore, calculating the contribution of the welding temperature and pressure parameters to void formation includes: The preprocessed temperature characteristic values With void defect characteristic index The correlation calculation is shown in the following formula: in, The characteristic values ​​of the temperature sequence are calculated as follows: in, The total number of samples, Let k be the temperature feature value of the k-th sample. This is the average of the temperature characteristic values ​​of all samples. Let be the void defect characteristic index of the k-th sample. The average value of the void defect characteristic index for all samples. This represents the total welding cycle time. Let be the temperature value of the k-th sample at time t. For time-series weighting functions, This is the temperature change sensitivity coefficient. The rate of temperature change; The correlation calculation for the pressure parameters is similar.

[0010] Furthermore, when the correlation coefficient of the welding temperature parameter or the pressure parameter is greater than a preset threshold, a boundary value for anomaly detection is constructed.

[0011] Furthermore, the boundary value is calculated by clustering the welding temperature parameter and the pressure parameter using a clustering algorithm to discover different types of abnormal patterns, obtain the cluster centers and intra-cluster distribution characteristics, thereby determining the abnormal boundaries of the parameters and obtaining the boundary value.

[0012] Furthermore, matching the corresponding welding temperature and pressure parameters with the clustering boundary values ​​includes: The feature values ​​of the welding temperature parameter and the pressure parameter are extracted respectively; If the temperature characteristic value within a certain time window exceeds the corresponding cluster boundary value, the deviation of the characteristic value from the nearest cluster center is calculated, and a deviation index is generated. The abnormal process parameter combinations and their specific time periods are determined based on the degree of deviation.

[0013] Furthermore, the deviation index is calculated by determining the Mahalanobis distance from the current feature value to the cluster center and the range of the covariance ellipse, quantifying the degree of deviation between the feature value and the nearest cluster center, and generating a normalized deviation index based on the exponential weighted accumulation of significant deviation feature components and time windows.

[0014] According to another aspect of the present invention, a diode package defect tracing system based on correlation analysis is provided, comprising: The data acquisition module is used to acquire process parameters during diode packaging and corresponding X-ray imaging images, and to establish a database of the correspondence between the process parameters and the X-ray imaging images. The feature decomposition module is used to perform Fourier transform processing on the X-ray imaging image, extract the frequency domain feature values ​​of the image, obtain the high-frequency and low-frequency components of the frequency domain feature values ​​through wavelet decomposition, and use the ratio of the high-frequency component to the low-frequency component as the void defect feature index. The process matching module is used to calculate the contribution of welding temperature parameters and pressure parameters to void formation based on the process parameters and the void defect characteristic index, identify process parameters with correlation coefficients greater than a preset threshold as key influencing factors, and construct K-means clustering boundary values ​​based on the historical distribution of the key influencing factors. The abnormal combination determination module is used to extract the void defect characteristic index when a new diode package void defect is detected, and match the corresponding welding temperature parameters and pressure parameters with the cluster boundary value to determine the abnormal process parameter combination that causes the void defect.

[0015] Compared with existing technologies, the diode packaging defect tracing method and system provided by this invention, based on correlation analysis, establishes a correlation model between process parameters and X-ray imaging images during the diode packaging process. It extracts frequency domain feature values ​​from the X-ray images and performs correlation analysis with the process parameters to determine the key abnormal process parameter combinations that cause void defects. This comprehensively improves the accuracy and reliability of defect tracing, helps to quickly locate abnormal process parameters, optimize packaging process conditions, and effectively reduce the incidence of void defects, thereby improving the quality and performance of diode packaging. It can also, to some extent, solve the problem of accurately locating the root process parameters of void defects caused by mismatched soldering temperature and pressure. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings: Figure 1 This is a flowchart of a diode packaging defect tracing method based on correlation analysis according to an embodiment of the present invention.

[0017] Figure 2 This is a flowchart illustrating the matching of process parameters with clustering boundary values ​​in a diode packaging defect tracing method based on correlation analysis according to an embodiment of the present invention. Detailed Implementation

[0018] Hereinafter, exemplary embodiments according to the present invention will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of the present invention, and not all embodiments of the present invention. It should be understood that the present invention is not limited to the exemplary embodiments described herein.

[0019] Figure 1 This is a flowchart of a diode packaging defect tracing method based on correlation analysis according to an embodiment of the present invention. Figure 1 As shown, the diode packaging defect tracing method based on correlation analysis includes: S1: Collect welding temperature parameters, pressure parameters, time parameters, and corresponding X-ray imaging images during the diode packaging process, and establish a database of the correspondence between the process parameters and the X-ray imaging images; On the diode packaging production line, multiple sets of process parameter data are first acquired during the diode packaging process. Specifically, this includes: real-time recording of welding temperature parameters using temperature sensors at the welding station (temperature acquisition frequency 100Hz, acquisition duration covering the entire welding process); simultaneous acquisition of welding pressure parameters using pressure sensors (pressure acquisition frequency 100Hz); and recording of welding time parameters using a high-precision timer. For each diode under test, immediately after the welding process is completed, it is imaged using a microfocus X-ray imaging system. The X-ray tube voltage is set to 80kV, the current to 100μA, and the exposure time to 200ms, obtaining a 1024×1024 pixel digital X-ray image. The welding temperature, pressure, and time parameters, along with the corresponding X-ray images obtained in the above process, are associated and stored according to production batch, time sequence, and workpiece number, establishing a structured database containing the correspondence between process parameters and images. This database enables one-to-one traceability of process parameters and packaging quality, laying the data foundation for subsequent defect feature extraction and correlation analysis.

[0020] S2: Perform Fourier transform processing on the X-ray imaging image to extract the frequency domain feature values ​​of the image. Obtain the high-frequency and low-frequency components of the frequency domain feature values ​​through wavelet decomposition. Use the ratio of the high-frequency component to the low-frequency component as the void defect feature index. First, the acquired X-ray images are preprocessed, including image denoising and contrast enhancement. Specifically, median filtering is used to remove salt-and-pepper noise and random noise from the images, and histogram equalization is used to improve image contrast and enhance the gray-level difference between the cavity / defect region and the normal region. The preprocessed images are more suitable for subsequent feature extraction.

[0021] Then, a two-dimensional Fast Fourier Transform is performed on the preprocessed image to transform it from the spatial domain to the frequency domain. In the frequency domain, the boundary and internal structural features of void defects are reflected in different frequency components. To enhance the expressive power of frequency domain features, an adaptive weight enhancement process is introduced. This weighting method dynamically adjusts the weights according to the importance of frequency components, assigning greater weights to high-frequency regions to highlight defect boundary features and less weights to low-frequency regions to preserve overall structural information. This adaptive weighting design can effectively enhance the feature representation of void defects in the frequency domain. Let the original image matrix be... Its frequency domain eigenvalue matrix after Fourier transform The calculation is as follows: in, and These represent the number of rows and columns of the image, respectively. and For frequency domain coordinates, Spatial domain coordinates, The imaginary unit is used. To enhance the expressive power of frequency domain features, an adaptive weighting coefficient α is introduced to weight the spectrum: in, This is an adaptive weight coefficient matrix. The adjustable parameter is used to control the weight distribution, and this weight design can highlight the frequency domain characteristics of void defects.

[0022] Next, the weighted frequency domain features are subjected to three-level Haar wavelet decomposition to obtain the corresponding high-frequency sub-bands. , , (i=1,2,3) and low-frequency subband ( The Haar wavelet basis was chosen because of its excellent edge-preserving properties, making it suitable for detecting boundary information of void defects. In each decomposition layer, one low-frequency subband and three high-frequency subbands are obtained. The low-frequency subband reflects the approximate components of the image, containing overall energy distribution information of the void region; the high-frequency subbands contain detailed information in the horizontal, vertical, and diagonal directions, reflecting the boundary features and texture variations of void defects.

[0023] Furthermore, when extracting high-frequency components, considering the different contributions of different decomposition levels to the void features, a multi-scale weighted approach is used to calculate the high-frequency component values. Specifically, the energy of the three high-frequency subbands obtained from each decomposition level is calculated, and different weight coefficients are assigned according to the decomposition level. Deeper decomposition results have greater weight because deeper decomposition can capture larger-scale structural features and is more meaningful for the overall characterization of void defects. High-frequency component values The calculation is as follows: in, Let be the weight coefficients of the i-th layer, satisfying: in, For the i-th level wavelet decomposition, the horizontal high-frequency subband is... Let be the vertical high-frequency subband of the i-th layer wavelet decomposition. For the diagonal high-frequency subband of the i-th level wavelet decomposition, The wavelet decomposition level is denoted as . This is the count variable for the summation term.

[0024] For the extraction of low-frequency components, the focus is primarily on the energy distribution characteristics of the last low-frequency sub-band. Considering that the void region exhibits localized energy variations within the low-frequency sub-band, a standard deviation term is introduced to correct for the low-frequency components. This correction better reflects the energy difference between the void region and the surrounding normal region. Therefore, the low-frequency components... The calculation is as follows: in, for The standard deviation of the sub-band As an adjustment coefficient, this term is introduced to better characterize the overall features of the void region.

[0025] After obtaining the high-frequency and low-frequency component values, their ratio is calculated as the initial feature index. To improve the discriminative power of the feature index, dynamic changes between adjacent image frames are also considered. Specifically, the rate of change of high-frequency and low-frequency components in continuously acquired images is calculated, and this dynamic change information is integrated into the final feature index calculation. This combination of static and dynamic features gives the obtained cavity defect feature index a stronger discriminative power. The final cavity defect feature index... The calculation is as follows: in, and These represent the rates of change of high-frequency and low-frequency components between adjacent image frames, respectively. This serves as a dynamic adjustment factor. This design combines static and dynamic features, improving the discriminative power of the feature index.

[0026] It is worth noting that although the above description outlines the calculation process for the void defect characteristic index, the specific implementation details can be illustrated through the following example: For example, a typical X-ray imaging image of a diode package yields a frequency domain eigenvalue matrix after Fourier transform. In the calculation of the adaptive weighting coefficient α, the weight distribution exhibits an increasing trend from the center to the edges: the weight is smaller in the central frequency domain region and larger in the edge region. This weight distribution design fully considers the energy distribution characteristics of void defects in the frequency domain, which is beneficial for highlighting the boundary information of the defects.

[0027] In wavelet decomposition, taking a real-world example of a cavity defect, after three layers of decomposition, sub-bands at different scales contain feature information at different levels. By appropriately setting the weighting coefficients, the calculation results are made to focus more on the decomposition layers containing significant cavity features, because the wavelet coefficients of deeper layers can usually better characterize the structural features of cavities.

[0028] For calculating low-frequency component values, when voids appear in the diode package, the standard deviation of the LL3 subband is significantly higher than that of the normal package area. Introducing an adjustment coefficient γ can effectively amplify this difference, making the feature more discriminative. In dynamic features, when the rate of change of high-frequency components ΔH / ΔL between two adjacent frames exceeds a specific threshold, it often indicates the presence of significant void defects in that area. Experimental verification shows that package quality can be divided into three levels based on the numerical range of the final calculated void defect characteristic index D_index: significant void defects, minor voids, and good packaging. This grading standard can provide a quantitative basis for quality control on the production line.

[0029] Ultimately, the goal is to accurately identify and characterize void defects in diode packages, providing reliable characteristic indicators for subsequent defect tracing analysis. These parameter settings may need to be adjusted appropriately based on specific production line conditions and quality requirements, but the core principles of the calculation method remain unchanged. In practical applications, the optimal parameter configuration can be determined through sample testing to achieve the best defect detection results.

[0030] S3: Based on the process parameters and the void defect characteristic index, the contribution of welding temperature parameters and pressure parameters to void formation is calculated using the Pearson correlation coefficient. Process parameters with a correlation coefficient greater than 0.8 are identified as key influencing factors. K-means clustering boundary values ​​are constructed based on the historical distribution of the key influencing factors. Before conducting correlation analysis, feature extraction and preprocessing are required for the collected welding temperature parameter sequence T_k(t) and pressure parameter sequence P_k(t). For each sample k, the temperature parameter sequence contains temperature change data throughout the entire welding cycle τ. Considering the significant impact of the temperature change rate on void formation during welding, a time-series weighting function ω(t) is introduced. The design principle of this weighting function is that a large temperature change rate indicates a possible significant material state change during this period, thus assigning a larger weight value. The T_k^* obtained through integration retains both the absolute temperature value information and highlights the contribution of the drastic temperature change interval. Similarly, a similar processing method is used for the pressure parameter sequence. The pressure change rate is weighted using the ν(t) function, focusing on the impact of pressure abrupt change intervals on void formation. This preprocessing method better reflects the dynamic characteristics of parameter changes compared to simple mean or maximum value statistics.

[0031] This invention proposes an improved correlation analysis method based on the traditional Pearson correlation coefficient. First, the preprocessed temperature characteristic value T_k^* is correlated with the void defect characteristic index D_index_k. The calculation process not only considers the degree of linear correlation but also enhances the influence of anomalous samples by introducing sample weights. Specifically, when the characteristic index of a sample deviates significantly from the average, it is given a larger weight when calculating the correlation coefficient, thus better capturing the parameter characteristics that lead to severe defects. The same strategy is used for the correlation analysis of pressure parameters. More specifically, the correlation coefficient of pressure parameters is calculated as follows: in, The characteristic values ​​of the temperature sequence are calculated as follows: in, The total number of samples, Let k be the temperature feature value of the k-th sample. This is the average of the temperature characteristic values ​​of all samples. Let be the void defect characteristic index of the k-th sample. The average value of the void defect characteristic index for all samples. This represents the total welding cycle time. Let be the temperature value of the k-th sample at time t. For time-series weighting functions, This is the temperature change sensitivity coefficient. This represents the rate of temperature change.

[0032] Similarly, the correlation coefficient R_P of the pressure parameter is calculated as follows: in, The calculation uses a similar time-weighted method: in, Let be the pressure value of the k-th sample at time t. The time-series weighting function for pressure is... This is the pressure change sensitivity coefficient. The rate of change of pressure, These are characteristic values ​​of the pressure parameter.

[0033] For process parameters identified as key influencing factors (correlation coefficient greater than 0.8), boundary values ​​for anomaly detection need to be constructed. An improved K-means clustering algorithm is adopted, its innovation lying in the adaptive update mechanism of cluster centers. Traditional K-means algorithms assign the same weight to all samples when updating cluster centers, while our proposed exponential decay weight scheme considers the influence of distance from the sample to the cluster center. Samples closer to the center have a larger weight; this mechanism improves the stability and accuracy of clustering. Let the set of normal samples in historical data be X_n, and the set of abnormal samples be X_a. The cluster center update strategy is as follows: in, For the updated i-th cluster center, For the i-th cluster center before the update, The learning rate (recommended value range is 0.01-0.1). Let be the sample points belonging to the i-th class. Let i be the set of samples of class i. This is the distance attenuation coefficient (recommended value range: 0.1-1.0). is the Euclidean distance from the sample point to the cluster center.

[0034] Furthermore, in the calculation of boundary values, not only is the intra-class dispersion (characterized by the standard deviation σ_i) considered, but the influence of inter-class distance is also introduced. When the cluster centers of different classes are close together, the boundary value is appropriately narrowed by adjusting the factor θ to reduce false positives; when the inter-class distance is large, the boundary value can be appropriately widened to improve the detection sensitivity. This adaptive boundary value determination method can better adapt to the dynamic changes in the distribution of process parameters. As shown in the following equation: in, Let be the boundary value of the i-th class. Let i be the cluster center of the i-th class. Let be the standard deviation of the i-th class of samples. This is the boundary adjustment coefficient (recommended value range: 0.5-2.0). To minimize the inter-class distance, This represents the maximum inter-class distance.

[0035] S4: When a new diode package void defect is detected, its void defect characteristic index is extracted, and the corresponding welding temperature parameters and pressure parameters are matched with the cluster boundary value to determine the abnormal process parameter combination that causes the void defect.

[0036] First, feature extraction was performed on newly detected diode package void defect samples. Imaging images of these samples were acquired using an X-ray imaging system, and the corresponding complete welding temperature and pressure parameter sequences were recorded simultaneously. During image acquisition, imaging parameters such as X-ray tube voltage, current, and exposure time were ensured to remain consistent with historical samples to guarantee the comparability of feature extraction. The acquired images underwent preprocessing, including noise reduction and contrast enhancement, to improve image quality.

[0037] Next, the void defect characteristic index of the sample is calculated. The frequency domain features of the image are obtained using the aforementioned Fourier transform and wavelet decomposition methods. Based on the frequency domain features, high-frequency and low-frequency components are extracted separately. The high-frequency components mainly reflect the detailed features of the void edges and internal structure, while the low-frequency components reflect the overall morphological features of the void. The two components are weighted and combined to obtain an index value that comprehensively characterizes the void features. The larger the index value, the higher the severity of the void defect.

[0038] After obtaining the temperature and pressure parameter sequences of the diode package samples, the parameter sequences are first segmented into time windows, each corresponding to a different stage of the welding process. For temperature parameters, feature values ​​such as the heating rate in the preheating stage, the peak temperature and holding time in the welding stage, and the cooling rate in the cooling stage are extracted. For pressure parameters, feature values ​​such as the pressure rise rate in the pressurization stage, the pressure stability and pressure fluctuation range in the welding stage, and the pressure release rate in the depressurization stage are extracted. After extracting the feature values, these feature values ​​are matched with the boundary values ​​obtained by K-means clustering in advance at multiple levels. If the temperature feature value in a certain time window exceeds the corresponding cluster boundary value, the deviation of the feature value from the nearest cluster center is calculated, and the distribution of pressure feature values ​​in that time window is evaluated. If both temperature and pressure feature values ​​are found to be abnormal, the coupling relationship between the two is further analyzed, and a comprehensive anomaly score is calculated. When the temperature feature value is within the boundary value range but close to the boundary, a comprehensive judgment needs to be made in conjunction with the stability of the pressure parameter in that time window. The feature value matching results of different process stages are fused using a weighted method, with a focus on the parameter matching situation in the welding stage. If the temperature or pressure parameters at a certain process stage deviate significantly, that stage is marked as a critical anomaly, and the corresponding parameter combination is recorded. Through this multi-dimensional matching analysis, the abnormal process parameter combinations that lead to the formation of void defects and the specific time periods in which they occur are ultimately determined, providing a precise basis for subsequent process parameter optimization.

[0039] like Figure 2When calculating the deviation of a feature value from its nearest cluster center, the temperature and pressure parameters within the current time window are first extracted. Specifically, this includes the temperature value, temperature change rate, pressure value, and pressure change rate at each sampling point. Simultaneously, the temperature fluctuation range, pressure fluctuation range, mean temperature, and mean pressure within the time window are calculated. For each cluster center, the Mahalanobis distance from the current feature value to that cluster center is calculated, standardized using the covariance matrix of historical samples. After calculating the distances to all cluster centers, the cluster center with the smallest distance is selected as the reference point. For the selected nearest cluster center, it is first determined whether the current feature value falls within the covariance ellipse of that cluster. When the feature value is inside the covariance ellipse, its projected distance along the principal axis of the ellipse is calculated, and the deviation direction is determined based on the projection position. When the feature value is outside the covariance ellipse, the shortest distance from the feature value to the ellipse boundary is used as the basic deviation. Weighting coefficients are set for different feature components when quantifying the degree of deviation. When the deviation of a feature component exceeds twice the standard deviation of the cluster, that feature component is marked as significantly deviating. For cases where both temperature and pressure features deviate simultaneously, the weighted sum of the squares of their deviations is used as the coupling deviation degree. When significant deviations occur in three or more consecutive time windows, the deviations of these windows are exponentially weighted and accumulated, with a decay coefficient set to 0.8. Finally, a normalized deviation index is output, ranging from 0 to 1, with a higher value indicating a more severe deviation. This deviation index will serve as an important basis for judging the degree of abnormality of process parameters and will be used for subsequent identification of abnormal process parameter combinations.

[0040] In summary, the diode packaging defect tracing method based on correlation analysis, as described in this invention, is explained. By establishing a correlation model between process parameters and X-ray imaging images during the diode packaging process, extracting frequency domain feature values ​​from the X-ray images, and combining this with process parameters for correlation analysis, the key abnormal process parameter combinations causing void defects are identified. This comprehensively improves the accuracy and reliability of defect tracing, helps to quickly locate abnormal process parameters, optimize packaging process conditions, thereby effectively reducing the incidence of void defects, improving the quality and performance of diode packaging, and to some extent solving the problem of accurately locating the root process parameters of void defects caused by mismatched soldering temperature and pressure.

[0041] According to another aspect of the present invention, a diode package defect tracing system based on correlation analysis is provided, comprising: The data acquisition module is used to acquire process parameters during diode packaging and corresponding X-ray imaging images, and to establish a database of the correspondence between the process parameters and the X-ray imaging images. The feature decomposition module is used to perform Fourier transform processing on the X-ray imaging image, extract the frequency domain feature values ​​of the image, obtain the high-frequency and low-frequency components of the frequency domain feature values ​​through wavelet decomposition, and use the ratio of the high-frequency component to the low-frequency component as the void defect feature index. The process matching module is used to calculate the contribution of welding temperature parameters and pressure parameters to void formation based on the process parameters and the void defect characteristic index, identify process parameters with correlation coefficients greater than a preset threshold as key influencing factors, and construct K-means clustering boundary values ​​based on the historical distribution of the key influencing factors. The abnormal combination determination module is used to extract the void defect characteristic index when a new diode package void defect is detected, and match the corresponding welding temperature parameters and pressure parameters with the cluster boundary value to determine the abnormal process parameter combination that causes the void defect.

[0042] Here, those skilled in the art will understand that the specific operations of each step in the aforementioned diode packaging defect tracing system based on correlation analysis have been referenced above. Figure 1 and Figure 2 The method for tracing the source of diode package defects based on correlation analysis has been described in detail, and therefore, its repeated description will be omitted.

[0043] In summary, the diode packaging defect tracing system based on correlation analysis, as described in this invention, is explained. By establishing a correlation model between process parameters and X-ray imaging images during the diode packaging process, extracting frequency domain feature values ​​from the X-ray images, and performing correlation analysis with the process parameters, the system identifies the key abnormal process parameter combinations that cause void defects. This comprehensively improves the accuracy and reliability of defect tracing, helps to quickly locate abnormal process parameters, optimize packaging process conditions, effectively reduce the incidence of void defects, and improve the quality and performance of diode packaging. It can also, to some extent, solve the problem of accurately locating the root cause process parameters of void defects caused by mismatched soldering temperature and pressure.

Claims

1. A method for tracing the source of diode packaging defects based on correlation analysis, characterized in that, include: Collect process parameters during diode packaging and corresponding X-ray imaging images, and establish a database of the correspondence between the process parameters and X-ray imaging images; The X-ray imaging image is subjected to Fourier transform processing to extract the frequency domain feature values ​​of the image. The high-frequency and low-frequency components of the frequency domain feature values ​​are obtained by wavelet decomposition. The ratio of the high-frequency component to the low-frequency component is used as the void defect feature index. Based on the process parameters and the void defect characteristic index, the contribution of welding temperature parameters and pressure parameters to void formation is calculated. Process parameters with correlation coefficients greater than a preset threshold are identified as key influencing factors. K-means clustering boundary values ​​are constructed based on the historical distribution of the key influencing factors. When a new diode package void defect is detected, its void defect characteristic index is extracted, and the corresponding welding temperature parameters and pressure parameters are matched with the cluster boundary value to determine the abnormal process parameter combination that causes the void defect. Matching the corresponding welding temperature and pressure parameters with the clustering boundary values ​​includes: The sequence of welding temperature parameters and pressure parameters is divided into time windows, with each time window corresponding to a different stage of the welding process; Extract the feature values ​​of temperature parameters and pressure parameters within each time window, and perform multi-level matching between the extracted feature values ​​and the cluster boundary values; If the temperature characteristic value in the current time window exceeds the corresponding cluster boundary value, the deviation of the characteristic value from the nearest cluster center is calculated, and the distribution of the pressure characteristic value in the time window is evaluated simultaneously. If both temperature and pressure characteristic values ​​are abnormal within the current time window, the coupling relationship between the two is analyzed, and a comprehensive anomaly score is calculated. If the temperature characteristic value within the current time window is within the boundary value range but close to the boundary, a comprehensive judgment is made based on the stability of the pressure parameter within that time window. The feature value matching results of each process stage are weighted and fused. Process stages with significant deviations are marked as key abnormal stages, and the corresponding abnormal process parameter combinations and occurrence time periods are recorded.

2. The diode packaging defect tracing method based on correlation analysis according to claim 1, characterized in that, Calculating the void defect characteristic index includes: Calculate the frequency domain eigenvalue matrix after Fourier transform. And perform weighted processing; The weighted frequency domain features are subjected to three-level Haar wavelet decomposition to obtain the corresponding high-frequency sub-band and low-frequency sub-band. Energy calculations are performed based on the high-frequency subband and the low-frequency subband respectively to obtain the high-frequency component value and the low-frequency component value; The ratio of the high-frequency component value to the low-frequency component value is calculated, and the dynamic change information is integrated to obtain the void defect characteristic index.

3. The diode packaging defect tracing method based on correlation analysis according to claim 2, characterized in that, The frequency domain eigenvalue matrix The calculation is shown in the following formula: in, and These represent the number of rows and columns of the image, respectively. and For frequency domain coordinates, Spatial domain coordinates, It is the imaginary unit.

4. The diode packaging defect tracing method based on correlation analysis according to claim 2, characterized in that, The dynamic change information is the rate of change of high-frequency and low-frequency components in continuously acquired images. The ratio of this rate of change to the high-frequency and low-frequency component values ​​is integrated with a dynamic adjustment factor to obtain the void defect characteristic index.

5. The diode packaging defect tracing method based on correlation analysis according to claim 1, characterized in that, The calculation of the contribution of the welding temperature and pressure parameters to void formation includes: The preprocessed temperature characteristic values With void defect characteristic index The correlation calculation is shown in the following formula: in, The characteristic values ​​of the temperature sequence are calculated as follows: in, The total number of samples, Let k be the temperature feature value of the k-th sample. This is the average of the temperature characteristic values ​​of all samples. Let be the void defect characteristic index of the k-th sample. The average value of the void defect characteristic index for all samples. This represents the total welding cycle time. Let be the temperature value of the k-th sample at time t. For time-series weighting functions, This is the temperature change sensitivity coefficient. The rate of temperature change; The correlation calculation for the pressure parameters is similar.

6. The diode packaging defect tracing method based on correlation analysis according to claim 5, characterized in that, When the correlation coefficient of the welding temperature parameter or the pressure parameter is greater than a preset threshold, a boundary value for anomaly detection is constructed.

7. The diode packaging defect tracing method based on correlation analysis according to claim 6, characterized in that, The boundary value is calculated by clustering the welding temperature parameter and the pressure parameter using a clustering algorithm to discover different types of abnormal patterns, obtain the cluster centers and intra-cluster distribution characteristics, thereby determining the abnormal boundaries of the parameters and obtaining the boundary value.

8. The diode packaging defect tracing system based on correlation analysis according to claim 1, characterized in that, include: Matching the corresponding welding temperature and pressure parameters with the clustering boundary values ​​includes: The feature values ​​of the welding temperature parameter and the pressure parameter are extracted respectively; If the temperature characteristic value within a certain time window exceeds the corresponding cluster boundary value, the deviation of the characteristic value from the nearest cluster center is calculated, and a deviation index is generated. The abnormal process parameter combinations and their specific time periods are determined based on the degree of deviation.

9. The diode packaging defect tracing method based on correlation analysis according to claim 8, characterized in that, The deviation index is calculated by determining the Mahalanobis distance from the current feature value to the cluster center and the range of the covariance ellipse, quantifying the degree of deviation between the feature value and the nearest cluster center, and generating a normalized deviation index based on the exponential weighted accumulation of significant deviation feature components and time windows.

10. A diode packaging defect tracing system based on correlation analysis, based on the diode packaging defect tracing method based on correlation analysis according to any one of claims 1 to 9, characterized in that, The data acquisition module is used to acquire process parameters during diode packaging and corresponding X-ray imaging images, and to establish a database of the correspondence between the process parameters and the X-ray imaging images. The feature decomposition module is used to perform Fourier transform processing on the X-ray imaging image, extract the frequency domain feature values ​​of the image, obtain the high-frequency and low-frequency components of the frequency domain feature values ​​through wavelet decomposition, and use the ratio of the high-frequency component to the low-frequency component as the void defect feature index. The process matching module is used to calculate the contribution of welding temperature parameters and pressure parameters to void formation based on the process parameters and the void defect characteristic index, identify process parameters with correlation coefficients greater than a preset threshold as key influencing factors, and construct K-means clustering boundary values ​​based on the historical distribution of the key influencing factors. The abnormal combination determination module is used to extract the void defect characteristic index when a new diode package void defect is detected, and match the corresponding welding temperature parameters and pressure parameters with the cluster boundary value to determine the abnormal process parameter combination that causes the void defect.

Images