A PCB resin plug hole defect detection and quality evaluation method based on multi-scale 3D feature fusion

Abstract

The application belongs to the technical field of PCB detection, and particularly relates to a PCB resin plug hole defect detection and quality evaluation method based on multi-scale 3D feature fusion, which adopts continuous spiral scanning to collect full-board 3D X-RAY data and reconstructs to generate three-dimensional space data, positions the target analysis area of each plug hole after pretreatment, extracts the surface topography features, internal structure features and interface combination features of each plug hole through multi-scale Gaussian pyramid decomposition and attention weighted fusion, and respectively identifies surface depression / convex, internal bubble / crack, copper-resin poor combination and missing plug / half plug / position offset defects. The application can realize one-time detection of all types of defects such as full-board surface depression, internal bubble, resin cracking, poor interface combination, missing plug, half plug and position offset, solves the efficiency bottleneck that traditional X-RAY can only be spot-checked and cannot cover the full board, and solves the blind area problem that AOI cannot detect internal defects.

Description

Technical Field

[0001] This invention belongs to the field of PCB inspection technology, specifically relating to a method for detecting and evaluating PCB resin plug defects based on multi-scale 3D feature fusion. Background Technology

[0002] In the high-end PCB manufacturing field, resin via plugging is a key process for achieving high-density interconnect (HDI), ensuring interlayer insulation, and improving product reliability. As PCBs evolve towards thinner and higher-density designs, via diameters are continuously shrinking, and the number of vias per board is surging, leading to increasingly stringent requirements for via filling quality. Core quality defects include not only visible resin depressions, protrusions, or foreign objects, but also hidden defects such as internal air bubbles, resin cracking, and poor adhesion between copper foil and resin. These internal defects can cause copper plating voids, interlayer short circuits, and even board failure during subsequent electroplating or high-temperature processes, making them one of the main causes of failure in high-end products.

[0003] Currently, among the mainstream inspection methods in the industry, traditional 2D AOI (Automated Optical Inspection) and 3D white light interferometry can only acquire two-dimensional or morphological information of the surface of plugged holes, completely failing to detect hidden defects such as internal bubbles and poor bonding, and also having a high false alarm rate. Traditional 3D X-ray (three-dimensional X-ray inspection), while capable of penetrating internal structures, often employs extremely time-consuming fixed-point tomographic scanning, failing to meet the full-board coverage and high-volume production inspection requirements. Manual microscope inspection, on the other hand, is extremely inefficient and highly subjective. In summary, existing technologies cannot simultaneously meet the stringent requirements of full-board coverage, micron-level accuracy, and three-dimensional quantification of internal defects. The industry urgently needs a fully automated solution that balances transmission inspection capabilities with production cycle time. Summary of the Invention

[0004] The purpose of this invention is to provide a method for detecting and evaluating PCB resin plug defects based on multi-scale 3D feature fusion, so as to solve the problem of full-board micron-level three-dimensional detection of hidden defects such as air bubbles, resin cracking, and poor copper-resin bonding in PCB resin plugs.

[0005] The present invention achieves the above objectives through the following technical solutions: Firstly, this invention proposes a method for detecting PCB resin via plugging defects based on multi-scale 3D feature fusion, applied to a 3D X-ray inspection system including an X-ray source, detector, and motion platform. The method includes the following steps: The X-ray source and detector are controlled to rotate synchronously, and the motion platform is controlled to drive the PCB board under test to move at a constant speed. The PCB board under test is continuously spiral scanned to collect the full board 3D X-ray data, which is then reconstructed to generate the full board three-dimensional spatial data. The three-dimensional spatial data is preprocessed with adaptive Gaussian denoising and contrast enhancement to obtain three-dimensional spatial data with enhanced material contrast, and the target analysis area where each plug hole is located is located by grayscale threshold segmentation. Multi-scale Gaussian pyramid decomposition of each target analysis region is performed and then fused with attention weighting to extract multi-dimensional features of each plug, including surface morphology features, internal structure features and copper-resin interface features. Based on the multidimensional features, surface defects, internal defects, and interface defects of each plug hole are identified and the corresponding defect parameters are quantified. Furthermore, based on the comparison between the designed hole positions and the actual hole positions detected from the three-dimensional spatial data, defects such as full-board leakage, partial plugging, and plugging hole position offset are identified.

[0006] Furthermore, the reconstruction to generate full-plate three-dimensional spatial data includes the following steps: The 3D X-RAY data is filtered and back-projected to reconstruct the initial three-dimensional volume data. An iterative reconstruction algorithm is used to optimize the initial three-dimensional volume data to generate full-plate three-dimensional spatial data.

[0007] Furthermore, the three-dimensional spatial data for enhanced material contrast is obtained through the following steps: An adaptive three-dimensional Gaussian filter is used to denoise the full-plate three-dimensional spatial data. The filter kernel size is adjusted according to the local variance of the voxels to obtain the denoised three-dimensional spatial data. The denoised 3D spatial data is divided into several 3D sub-blocks. Histogram transformation is performed on each 3D sub-block. Gray levels exceeding a set value in each 3D sub-block are clipped. The clipped parts are redistributed to the remaining gray levels in the histogram of each 3D sub-block. The contrast variation of each 3D sub-block is controlled to obtain the transformed 3D spatial data. Interpolation smoothing is performed on the boundaries between the three-dimensional sub-blocks in the transformed three-dimensional spatial data to obtain three-dimensional spatial data with enhanced material contrast.

[0008] Furthermore, the method of locating the target analysis region where each resin plug pore is located by segmenting and locating the three-dimensional spatial data based on the material contrast enhancement through grayscale thresholding includes the following steps: Based on the grayscale difference between copper foil and resin in X-ray attenuation coefficient, a grayscale threshold is set for the three-dimensional spatial data of the material with enhanced contrast. Regions with grayscale values ​​greater than the threshold are divided into copper foil regions, and regions with grayscale values ​​less than the threshold are divided into resin-filled regions. Based on the geometry of each copper foil region, identify the orifice location and wall extent of each via; Using the center of each through hole as the center, the radius as the set ratio of the hole diameter, and the plate thickness as the height, a three-dimensional sub-region containing the complete through hole structure is extracted from the three-dimensional spatial data as the target analysis area for each resin through hole.

[0009] Furthermore, the full-plate leakage and partial-plate defects are identified through the following steps: Read the coordinates of the designed holes on the PCB board under test; Based on the grayscale features of the copper holes in the three-dimensional spatial data with enhanced material contrast, the coordinates of all actual hole locations are detected. Perform nearest neighbor matching between the designed hole positions and the actual hole positions; If the designed hole position does not match the actual hole position, it is judged as a leakage plug defect; If the designed hole position matches the actual hole position, but the resin filling height in the actual hole position is less than the set percentage of the plate thickness, it is determined to be a partial plug defect.

[0010] Furthermore, the plug hole position offset defect is identified through the following steps: For successfully matched holes, the actual hole coordinates are compared with the corresponding designed hole coordinates; If the actual hole position coordinates are offset from the designed hole position coordinates by more than a set threshold, it is determined to be a hole position offset defect.

[0011] Furthermore, the step of performing multi-scale Gaussian pyramid decomposition on each target analysis region and then performing attention-weighted fusion to extract multidimensional features of each resin plug pore includes: Three-scale Gaussian pyramid decomposition was performed on each target analysis region to obtain the internal microstructure features at the original resolution scale, the interface features at a 2x downsampling scale, and the surface morphology features at a 4x downsampling scale. The internal microstructure features, interface features and surface morphology features are fused according to preset weights, wherein different fusion weights are set for internal defects and surface defects respectively, to obtain the fused features of each target analysis region. Based on the fusion features, multidimensional features of each plug hole are extracted, including surface morphology features, internal structure features, and interface bonding features. The surface morphology features include the height distribution of the plugged hole surface and the relative height parameter based on the reference plane; the internal structural features include microbubble regions, linear edges of microcracks, and grayscale anomalies within the resin region; and the interface bonding features include the grayscale gradient distribution and gradient anomaly regions at the interface between the copper foil and the resin.

[0012] Furthermore, the method for identifying and quantifying surface defects includes: Using the direction perpendicular to the PCB board surface as the Z-axis, the Z-axis coordinates of the resin surface in each plane position within the via filling area are obtained layer by layer to obtain the height distribution data of the via filling surface. The height data of the copper foil area around the orifice is fitted with a plane to obtain the reference plane; Calculate the height difference between the resin surface and the reference surface to obtain the depression depth or protrusion height.

[0013] Furthermore, the method for identifying and quantifying internal defects includes: Using the direction perpendicular to the PCB board surface as the Z-axis, extract multiple Z-axis slices with equal spacing along the hole axis with a set step size; Resin regions in each slice are segmented based on grayscale thresholding; A three-dimensional region growth algorithm is used to grow connected regions within the resin region where the gray value is lower than a set threshold, thereby obtaining the three-dimensional region of the bubble and calculating its equivalent diameter. A three-dimensional edge detection algorithm is used to extract the linear edge structure within the resin area to identify resin cracks.

[0014] Furthermore, the method for identifying and quantifying interface defects includes: Based on the grayscale gradient of the three-dimensional spatial data with enhanced material contrast, the interface between the copper foil and the resin is extracted. Calculate the grayscale gradient value at each location on the interface; When the grayscale gradient value of a certain area is lower than a set threshold, the area is determined to be a poorly bound area; Calculate the area of ​​the poorly bonded region and its proportion of the total interface area.

[0015] Secondly, the present invention proposes a PCB resin plugging defect detection system for implementing the above method, comprising: An X-ray source used to emit X-rays; A detector is used to receive X-rays that penetrate the PCB board under test and convert them into electrical signals. The motion platform is used to support the PCB board under test and drive it to move at a constant speed during continuous spiral scanning. The GPU parallel data processing unit is used to filter and back-project the 3D X-RAY data acquired by continuous spiral scanning, and combine iterative reconstruction to generate full-board three-dimensional spatial data. It is also used to perform adaptive Gaussian denoising and contrast enhancement preprocessing, multi-scale Gaussian pyramid decomposition and attention-weighted fusion, as well as defect identification and quantization on the three-dimensional spatial data.

[0016] Thirdly, this invention proposes a method for evaluating the quality of PCB resin via plugs based on the above-mentioned detection method, comprising the following steps: Obtain the defect identification results and corresponding defect quantification parameters for each plug hole on the PCB board under test; According to the preset grading standard, the severity level of each type of defect in each plug hole is classified. Quality is evaluated from three dimensions: board level, single hole level, and process level. The overall quality score of the board is calculated based on the evaluation results to determine whether the PCB board under test is qualified or unqualified. In the preset grading standard, defects are divided into fatal defects, serious defects, general defects and minor defects from high to low. Fatal defects include leakage, partial blockage and air bubbles exceeding the set size; serious defects include poor bonding area exceeding the set ratio and resin cracking; general defects include air bubbles within the set size range and dents within the set depth range; minor defects include air bubbles smaller than the set size, dents smaller than the set depth and surface foreign matter.

[0017] The beneficial effects of this invention are as follows: 1. This invention achieves efficient acquisition of three-dimensional spatial data across the entire board through continuous spiral scanning, breaking through the efficiency bottleneck of traditional fixed-point X-ray inspection, which can only inspect a few plugged holes. This allows for the acquisition of complete internal structural information for all plugged holes on the board, eliminating the risk of missing internal defects due to sampling in traditional inspections. Furthermore, adaptive denoising and contrast enhancement are applied to the three-dimensional spatial data, significantly improving the grayscale difference between the copper foil and resin. Additionally, a three-scale Gaussian pyramid decomposition combined with attention-weighted fusion is employed, simultaneously balancing the high resolution required for detecting minute defects with the morphological integrity required for identifying macroscopic defects. This enables the simultaneous detection of defects at multiple scales, such as microbubbles, microcracks, and poor bonding, in a single inspection, resolving the contradiction between the detection accuracy of defects of different sizes.

[0018] 2. This invention automatically matches the designed hole positions with the actual detection hole positions, eliminating the need for manual verification of design documents. It can detect global integrity defects such as incomplete plugging, partial plugging, and positional misalignment in a single step. Together with the aforementioned filling quality defect detection, it constitutes a complete multi-dimensional defect detection system. The three-dimensional region growth algorithm can extract bubble regions of arbitrary shapes, unrestricted by bubble morphology; three-dimensional edge detection can identify microcracks only a few micrometers wide, overcoming the limitation of traditional two-dimensional slicing analysis in tracking crack spatial orientation; and interface gradient analysis can accurately identify peeling and voids between copper foil and resin. These methods, based on three-dimensional spatial data, achieve a comprehensive quantitative evaluation of the surface morphology, internal density, and interface bonding state of each plugged hole. Attached Figure Description

[0019] Figure 1 This is a schematic flowchart of a PCB resin plugging defect detection method in one embodiment of the present invention; Figure 2This is a schematic diagram of another process of the PCB resin plugging defect detection method in one embodiment of the present invention; Figure 3 This is a system block diagram of a PCB resin plugging defect detection system according to one embodiment of the present invention. Detailed Implementation

[0020] The present application will now be described in further detail with reference to the accompanying drawings. It should be noted that the following specific embodiments are only used to further illustrate the present application and should not be construed as limiting the scope of protection of the present application. Those skilled in the art can make some non-essential improvements and adjustments to the present application based on the above application content.

[0021] It is worth noting that due to the extremely wide size range of resin plugging defects, from 3μm microbubbles to poor bonding at the hundred-micrometer scale, it is difficult to simultaneously achieve both high detection sensitivity for micro-defects and high morphological integrity for large-scale defects at a single scale. Therefore, this disclosure performs a three-scale Gaussian pyramid decomposition on the target analysis region of each plugging pore, obtaining internal microstructure features at the original resolution scale, interface features at a 2x downsampling scale, and surface morphology features at a 4x downsampling scale. The original scale retains the highest resolution to capture details such as microbubbles and microcracks, while the downsampling scale filters out local noise to highlight the gradual morphology of the interface and the overall surface undulations. Based on this, different fusion weights are set for internal and surface defects, and attention-weighted fusion is applied to the features at each scale. This allows the algorithm to adaptively highlight effective information and suppress redundant information according to the defect type, thereby achieving high-precision identification of both micrometer-scale and macroscopic defects in a single detection.

[0022] First Embodiment Please see Figure 1 and Figure 2 A method for detecting PCB resin via plugging defects based on multi-scale 3D feature fusion is proposed and applied to a 3D X-ray inspection system including an X-ray source, detector, and motion platform. The method includes the following steps: S1: Control the X-ray source and detector to rotate synchronously, control the motion platform to drive the PCB board under test to translate at a constant speed, and perform continuous spiral scanning on the PCB board under test to collect full-board 3D X-ray data, which is then reconstructed to generate full-board three-dimensional spatial data.

[0023] In this step, the X-ray source and detector are positioned on opposite sides of the PCB board under test and kept in a fixed relative position. They rotate synchronously around the normal direction of the PCB board, while the motion platform translates at a uniform speed horizontally, causing the X-ray trajectory on the PCB board to be helical. Because the X-ray source and detector use synchronous rotation combined with platform translation, the scanning trajectory continuously covers the entire board area, eliminating seams or blind spots.

[0024] The process involves filtering and backprojecting the collected 3D X-RAY data to obtain initial 3D volume data, and then using an iterative reconstruction algorithm to optimize the initial 3D volume data to generate full-panel 3D spatial data.

[0025] Filtered backprojection can quickly convert projection data into three-dimensional voxel grayscale values, while iterative reconstruction repeatedly compares and corrects the differences between the reconstructed image and the measured projection data, effectively suppressing the star-shaped artifacts and noise amplification problems common in the filtered backprojection algorithm. This results in the generated full-panel three-dimensional spatial data having a high signal-to-noise ratio and spatial resolution, with the smallest voxel size reaching 1μm.

[0026] S2: Perform adaptive Gaussian denoising and contrast enhancement preprocessing on the three-dimensional spatial data to obtain three-dimensional spatial data with enhanced material contrast, and locate the target analysis area where each plug hole is located by grayscale threshold segmentation.

[0027] In this embodiment, the core principle of denoising is to use a smaller filter kernel in areas with high local variance, such as hole wall boundaries and copper foil edges, to preserve edge details; and to use a larger filter kernel in flat areas with low local variance, such as the interior of resin fillers and substrate areas, to fully suppress noise. This adaptive strategy avoids edge blurring or insufficient denoising caused by fixed kernel size filtering. Contrast enhancement divides the denoised 3D spatial data into several 3D sub-blocks, performs histogram transformation on each sub-block, clips gray levels exceeding a set value within each sub-block, and redistributes the clipped portions to the remaining gray levels in the histogram of each sub-block, controlling the contrast variation of each sub-block to obtain the transformed 3D spatial data. Finally, interpolation smoothing is performed on the boundaries between the sub-blocks in the transformed 3D spatial data to obtain 3D spatial data with enhanced material contrast. This enhancement method effectively widens the gray level difference between the copper foil area and the resin area, making copper and resin, which originally had similar gray levels, present clearly separable gray level layers in the data, facilitating subsequent copper foil-resin segmentation based on gray level thresholds.

[0028] Accordingly, the specific implementation process of target analysis area localization is as follows: Based on the grayscale difference between copper foil and resin in X-ray attenuation coefficient, a grayscale threshold is set for the three-dimensional spatial data with enhanced material contrast; areas with grayscale values ​​greater than the threshold are divided into copper foil areas, and areas with grayscale values ​​less than the threshold are divided into resin-filled areas; according to the geometric shape of each copper foil area, the position of each via and the range of the via wall are identified; with the center of each via as the center, the set ratio of the via diameter as the radius, and the plate thickness as the height, a three-dimensional sub-region containing the complete via structure is extracted from the three-dimensional spatial data as the target analysis area for each resin via.

[0029] S3: Perform multi-scale Gaussian pyramid decomposition on each target analysis region and then perform attention-weighted fusion to extract multi-dimensional features of each plug, including surface morphology features, internal structural features, and copper-resin interface features.

[0030] Because resin plugging defects vary greatly in size, from 3μm microbubbles to poor bonding at the hundred-micrometer scale, it is difficult to simultaneously achieve both the detection sensitivity of small defects and the morphological integrity of large defects at a single scale. Therefore, this embodiment performs three-scale Gaussian pyramid decomposition on each target analysis region to obtain internal microstructure features at the original resolution scale, interface features at a 2x downsampling scale, and surface morphology features at a 4x downsampling scale. The original scale retains the highest spatial resolution, capable of capturing grayscale anomalies and edge details of small defects such as microbubbles and microcracks; the 2x downsampling scale filters out local noise through smoothing, highlighting the gradual morphology of the copper foil-resin interface; and the 4x downsampling scale further compresses the data volume, reflecting the overall undulations and macroscopic morphology of the plugging surface.

[0031] Specifically, internal microstructure features, interface features, and surface morphology features are fused according to preset weights. Different fusion weights are set for internal defects and surface defects to obtain fused features for each target analysis region. This attention weighting mechanism enables the algorithm to adaptively highlight effective information and suppress redundant information based on defect type. Based on the fused features, multidimensional features of each plug are extracted. These multidimensional features include surface morphology features, internal structure features, and interface bonding features. Surface morphology features include the height distribution of the plug surface and the relative height parameter based on the reference plane, reflecting the concavity and convexity of the resin surface; internal structure features include microbubble regions, linear edges of microcracks, and gray-level anomalies within the resin region, reflecting the compactness of the resin filler; interface bonding features include the gray-level gradient distribution and gradient anomaly regions at the copper foil-resin interface, reflecting the bonding quality between the copper foil and the resin.

[0032] S4: Based on multi-dimensional features, identify surface defects, internal defects, and interface defects of each plug hole and quantify the corresponding defect parameters; and, based on the comparison between the designed hole position and the actual hole position detected from the three-dimensional space data, identify full-board missing plug, half-plug, and plug hole position offset defects.

[0033] Specifically, this disclosure employs targeted methods to identify and quantify the various defects detected: The surface height distribution of the surface defects is obtained by extracting the Z-axis slice of the plug hole opening, and the reference surface is obtained by performing plane fitting on the copper foil area around the hole opening. The height difference between the resin surface and the reference surface is calculated to obtain the depression depth or protrusion height. Internal defects were identified by extracting equidistant multi-layer slices along the axial direction of the plug hole, segmenting the resin region based on grayscale thresholds, obtaining the three-dimensional region of the bubble and calculating its equivalent diameter using a three-dimensional region growth algorithm, and identifying resin cracks using a three-dimensional edge detection algorithm. Interface defects are identified by extracting the interface between the copper foil and resin based on the grayscale gradient of three-dimensional spatial data. Poorly bonded areas are identified by analyzing the grayscale gradient values ​​at various locations on the interface, and the proportion of these areas to the total interface area is calculated. These methods quantify defect parameters from three dimensions: surface morphology, internal structure, and interface bonding.

[0034] In a preferred embodiment, reconstructing and generating full-panel three-dimensional spatial data includes the following steps: S11: Filtered backprojection reconstruction of 3D X-RAY data to obtain initial three-dimensional volume data; filtered backprojection can quickly obtain the initial three-dimensional grayscale distribution by backprojecting the ray attenuation signal of each projection angle along the ray path to three-dimensional space and accumulating the projection data of all angles.

[0035] S12: An iterative reconstruction algorithm is used to optimize the initial 3D volume data and generate full-plate 3D spatial data. The iterative reconstruction uses the reconstruction result of filtered back projection as the initial estimate. The projection data corresponding to the current reconstructed image is calculated through orthographic projection and compared with the measured projection data. The 3D volume data is corrected round by round according to the difference between the two. This process is repeated until the difference is less than a set threshold.

[0036] In a preferred embodiment, the three-dimensional spatial data for enhanced material contrast is obtained through the following steps: S21: Adaptive 3D Gaussian filtering is used to denoise the entire board's 3D spatial data. The filter kernel size is adjusted according to the local variance of the voxels to obtain the denoised 3D spatial data. The local variance of a voxel is obtained by calculating the dispersion of the gray values ​​of each voxel in its 3D neighborhood. In areas with a large local gray standard deviation (such as hole wall boundaries and copper foil edges), it indicates that there is a significant gray change in the area. In this case, a smaller filter scale is used to preserve edge details. In flat areas with a small local gray standard deviation (such as inside the resin filler and substrate area), it indicates that the gray change in the area is gradual. In this case, a larger filter scale is used to fully suppress noise. The filter function is: ; in, This represents the coordinates of the voxel currently being processed in 3D space. To use the current voxels Three-dimensional neighborhood centered on Within, the offset of each neighboring voxel relative to the current voxel; Input voxel grayscale values, The filtered voxel gray values ​​are... The range is defined as the three-dimensional neighborhood centered on the current voxel, where N is the normalization coefficient. The adaptive filtering scale parameter is expressed as follows: ; in, Based on the basic filtering scale, This is the local variance adjustment coefficient. This represents the standard deviation of voxel gray values ​​within a local 3D neighborhood. In this formula, the local standard deviation... The larger the standard deviation, the smaller the filter scale σ, the weaker the smoothing effect, and the edges are preserved; the smaller the local standard deviation, the larger the filter scale σ, the stronger the smoothing effect, and the noise in flat areas is fully suppressed.

[0037] S22: Divide the denoised 3D spatial data into several 3D sub-blocks, perform histogram transformation on each 3D sub-block, clip gray levels exceeding a set value within each 3D sub-block, and redistribute the clipped portions to the remaining gray levels in the histogram of each 3D sub-block. Control the contrast variation of each 3D sub-block to obtain the transformed 3D spatial data. Through clipping and redistribution, the problem of excessive noise amplification in flat areas is effectively suppressed.

[0038] S23: Interpolation smoothing is performed on the boundaries between the three-dimensional sub-blocks in the transformed three-dimensional spatial data to obtain three-dimensional spatial data with enhanced material contrast. After interpolation smoothing, the grayscale transition between each sub-block is continuous and natural, and no boundary artifacts caused by independently processing each sub-block appear in the volume data. After the above processing, the grayscale difference between the resin and the copper foil is significantly improved.

[0039] In a preferred embodiment, based on three-dimensional spatial data with enhanced material contrast, the target analysis area where each resin plug pore is located is segmented and located using grayscale thresholding, including the following steps: S24: Based on the grayscale difference in X-ray attenuation coefficients between copper foil and resin, a grayscale threshold is set for the 3D spatial data with enhanced material contrast. Copper has a much higher density than resin, resulting in stronger X-ray absorption and appearing as a high grayscale region in the reconstructed 3D spatial data. Resin has a lower density and appears as a low grayscale region. This grayscale difference is further amplified after the material contrast is enhanced, allowing for effective separation of copper foil and resin by setting a single global grayscale threshold.

[0040] S25: Divide the area with a gray value greater than the threshold into a copper foil area, and the area with a gray value less than the threshold into a resin-filled area.

[0041] S26: Based on the geometric shape of each copper foil region, identify the position of the orifice and the range of the hole wall of each via; in the copper foil region, the hole wall of the via presents a continuous annular structure, and its geometric shape has a circular feature, which can be used to distinguish it from the copper foil on the board surface, and thus determine the position of each via.

[0042] S27: Using the center of each through hole as the center, the set ratio of the hole diameter as the radius, and the plate thickness as the height, a three-dimensional sub-region containing the complete through hole structure is extracted from the three-dimensional spatial data as the target analysis area for each resin through hole. The data volume obtained after extraction is much smaller than the whole plate data, and subsequent processing is only performed on each sub-region, which greatly reduces the amount of calculation.

[0043] In a preferred embodiment, full-plate leakage and partial-plate leakage defects are identified through the following steps: Read the coordinates of the designed holes on the PCB board under test; based on the grayscale features of the copper holes in the three-dimensional spatial data with enhanced material contrast, detect the coordinates of all actual holes; in the three-dimensional spatial data, the copper holes appear as tubular high grayscale areas that penetrate the thickness of the board. By extracting the copper foil area layer by layer along the thickness direction and tracking its spatial continuity, all actual holes can be accurately detected.

[0044] The design hole positions are matched with the actual hole positions using the nearest neighbor method. The nearest neighbor method finds the nearest design hole position for each actual hole position. If the distance between the two is less than the matching threshold, the match is considered successful; otherwise, it is considered an invalid match.

[0045] If the designed hole position does not match the actual hole position, it is judged as a missing hole defect; if the designed hole position matches the actual hole position but the resin filling height in the actual hole position is less than a set percentage of the board thickness, it is judged as a partial hole defect. The resin filling height is determined by detecting the continuous axial extension range of the resin area in the hole. The grayscale characteristics of the resin area in three-dimensional data are different from those of copper foil and air, which allows for accurate measurement of its filling depth.

[0046] In a preferred embodiment, the plug position misalignment defect is identified through the following steps: For successfully matched holes, the actual hole coordinates are compared with the corresponding designed hole coordinates; if the offset of the actual hole coordinates relative to the designed hole coordinates is greater than a set threshold, it is determined to be a hole position offset defect.

[0047] In a preferred embodiment, the target analysis regions are decomposed into multi-scale Gaussian pyramids and then fused using attention-weighted methods to extract multidimensional features of each resin plug pore, including: Each target analysis region is decomposed into a three-scale Gaussian pyramid to obtain the internal microstructure features at the original resolution scale, the interface features at a 2x downsampling scale, and the surface morphology features at a 4x downsampling scale. The Gaussian pyramid is constructed as follows: the original resolution data is used as the bottom layer of the pyramid. The bottom layer data is downsampled by 2x by deleting even-numbered rows and columns after Gaussian low-pass filtering. The second layer is obtained by repeating the above operation on the second layer to obtain the third layer, which is the 4x downsampling scale.

[0048] Understandably, the original resolution scale (e.g., 1 μm voxel size) retains the highest spatial frequency components in 3D data. Microbubbles and microcracks are typically only a few micrometers to tens of micrometers in size, and their edge and internal grayscale anomalies can be completely preserved at the original resolution. Downsampling reduces or even eliminates the grayscale contrast of these small defects, therefore, internal microstructural features must be extracted at the original scale. A 2x downsampling scale enlarges the voxel size to 2 μm, compressing the data volume to 1 / 8 of the original scale. High-frequency noise is effectively filtered out, and the gradient transition zone at the copper foil-resin interface exhibits a stable grayscale gradient at this scale, suitable for extracting interface features reflecting the bonding state. A 4x downsampling scale further enlarges the voxel size to 4 μm, compressing the data volume to 1 / 64 of the original scale. At this point, the macroscopic undulations and depressions on the plug surface become the main variable components in the data, and minute noise and local burrs are sufficiently suppressed, suitable for extracting surface morphology features.

[0049] Internal microstructure features, interface features, and surface morphology features are fused according to preset weights. Different fusion weights are set for internal defects and surface defects to obtain fused features for each target analysis region. The weight settings are determined based on the defect type: when internal defects need to be detected, the original scale features and interface features are given higher weights; when surface defects need to be detected, surface morphology features are given higher weights. Through weight adjustment, the fused features can adaptively highlight information related to the target defect type.

[0050] Based on the fusion characteristics, multidimensional features of each plug hole are extracted. These multidimensional features include surface morphology features, internal structure features, and interface bonding features. Surface morphology features include the height distribution of the plugged hole surface and the relative height parameter based on the reference plane; internal structural features include microbubble regions, linear edges of microcracks, and gray-scale anomalies within the resin region; and interface bonding features include the gray-scale gradient distribution and gradient anomaly regions at the interface between the copper foil and the resin.

[0051] In a preferred embodiment, the method for identifying and quantifying surface defects includes: Using the direction perpendicular to the PCB board surface as the Z-axis, the target analysis area containing the plugged via is scanned layer by layer along the Z-axis. In each layer's XY plane slice, the boundary position of the resin surface in that layer is located. The Z-axis coordinate value corresponding to the upper surface of the resin at each (x,y) coordinate is obtained through boundary tracing. The Z-values ​​corresponding to all (x,y) positions are summarized to obtain the three-dimensional height distribution data of the plugged via surface. ,in This represents the Z-axis height value at the coordinate (x, y) position within the plane of the orifice.

[0052] A reference plane is obtained by performing plane fitting on the height data of the copper foil area around the aperture. Since the PCB board may experience overall warping during manufacturing and inspection, directly using the original height data for judgment will introduce systematic bias. The copper foil area has uniform thickness and a flat surface, and its height data can reflect the warping pattern of the board surface. The reference plane is obtained by performing plane fitting on the height data of the copper foil area using the least squares method. This represents the ideal reference height for the location of the plug hole. The coefficient is... , c is determined by minimizing the sum of squared deviations between all height data within the copper foil region and the fitted plane.

[0053] Calculate the height difference between the resin surface and the reference surface to obtain the depression depth or protrusion height; using the reference surface as a reference, the portion of the resin surface lower than the reference surface is the depression, and the portion higher than the reference surface is the protrusion. Depression Depth That is, the difference between the reference plane and the lowest point of the resin surface; the protrusion height. This refers to the difference between the highest point on the resin surface and the reference surface. Additionally, the average roughness of the resin surface can also be calculated. This is used to quantify the overall smoothness of a surface. Where N is the number of surface points involved in the calculation. This represents the Z-axis height of the resin surface at its coordinate position within the orifice plane. This represents the average height of the resin surface. The above relative measurement method eliminates the influence of board warpage on the measurement results.

[0054] Defects are determined based on the calculated parameters: when the depression depth is greater than the set threshold, it is determined to be a depression defect; when the protrusion height is greater than the set threshold, it is determined to be a protrusion defect; when the surface roughness exceeds the set threshold and there is a sudden change in local height, it is determined to be a surface scratch or foreign object defect based on the grayscale characteristics of the area.

[0055] In a preferred embodiment, the method for identifying and quantifying internal defects includes: Using the direction perpendicular to the PCB board surface as the Z-axis, extract multiple Z-axis slices with equal spacing along the hole axis with a set step size; The resin region in each slice is segmented based on grayscale threshold.

[0056] A 3D region growing algorithm is used to grow connected regions within the resin area where the grayscale value is below a set threshold, obtaining the 3D region of the bubble and calculating its equivalent diameter. Region growing starts from the seed voxel that meets the grayscale condition and expands to adjacent voxels, grouping all connected low-grayscale voxels into the same region. This method has no preset requirements on the bubble shape and can accurately extract bubbles of any irregular shape. The equivalent diameter is obtained by calculating the volume of the bubble region and converting it into the diameter of a sphere of equal volume.

[0057] A 3D edge detection algorithm is used to extract linear edge structures within the resin region to identify resin cracks. 3D edge detection calculates the grayscale gradient direction and amplitude of each voxel in 3D space, extracts voxel points with gradient amplitudes exceeding a set threshold, and connects these continuously arranged edge points into edge lines. When the length of an edge line exceeds a set value, its width is less than a set value, and it exhibits a linear direction, it is identified as a crack.

[0058] In a preferred embodiment, the method for identifying and quantifying copper-resin interface defects includes: Based on the grayscale gradient of 3D spatial data for enhanced material contrast, the interface between copper foil and resin is extracted. The X-ray attenuation coefficients of copper and resin differ significantly, and a significant jump in grayscale value occurs at their interface. This jump manifests as a local maximum of the grayscale gradient in 3D space. The grayscale gradient at the interface location is then analyzed. The result is obtained by calculating the square root of the sum of the squares of the grayscale change rates in each direction: ; in, For position grayscale gradient magnitude at that location This represents the voxel grayscale value. These are the partial derivatives of the voxel grayscale values ​​along the x, y, and z directions, respectively. By locating the location of the maximum gradient along the direction perpendicular to the interface, the spatial location of the interface can be determined.

[0059] Calculate the grayscale gradient values ​​at each location on the interface. When the grayscale gradient value of a certain area is lower than a set threshold and the continuous length of that area along the interface direction exceeds a set dimension, that area is determined to be a poorly bonded area. If the copper foil and resin are well bonded, the grayscale gradient at the interface should remain high and continuous; if there is peeling or voids, the material difference between the two sides of the interface disappears or weakens, and the grayscale gradient value decreases accordingly.

[0060] Calculate the area of ​​the poorly bonded region And calculate its area relative to the entire interface. proportion Defects are determined based on this ratio: when When the ratio is greater than or equal to the first set ratio, it is judged as a serious bonding defect; when When the ratio is between the first and second set ratios, it is judged as a general poor bonding defect; when If the ratio is less than the second set ratio, it is judged as a slight bonding defect.

[0061] Second Embodiment Please see Figure 3 A PCB resin plugging defect detection system for implementing the detection method in the first embodiment includes an X-ray source, a detector, a motion platform, and a GPU parallel data processing unit, specifically: The X-ray source is used to emit X-rays; preferably, the X-ray source is an open microfocus X-ray tube with a focal size ≤5μm, which can achieve high spatial resolution with a smaller focal size.

[0062] The detector is used to receive X-rays that have penetrated the PCB board under test and convert them into electrical signals; preferably, the detector is a linear array detector, which can distinguish minute differences in grayscale.

[0063] The motion platform is used to support the PCB board under test and drive it to move at a constant speed during continuous spiral scanning; preferably, the motion platform adopts an air-floating support structure to eliminate friction and vibration caused by mechanical contact.

[0064] The GPU parallel data processing unit is used to filter and back-project the 3D X-RAY data acquired by continuous spiral scanning, and combine iterative reconstruction to generate full-board 3D spatial data. It is also used for adaptive Gaussian denoising and contrast enhancement preprocessing, multi-scale Gaussian pyramid decomposition and attention-weighted fusion, as well as defect identification and quantization of the 3D spatial data.

[0065] Due to the massive amount of 3D spatial data across the entire board, a single scan can generate several gigabytes or even tens of gigabytes of data, requiring extensive voxel computation for reconstruction and subsequent processing. The GPU parallel data processing unit employs a multi-GPU architecture, dividing the 3D volumetric data into several sub-blocks based on spatial location. Each GPU is responsible for processing different sub-blocks, significantly reducing data processing time. The various modules of this system work together to achieve fully automated detection from X-ray emission, signal acquisition, data reconstruction to defect identification.

[0066] Specifically, the GPU parallel data processing unit includes: The data reconstruction module is used to perform filtered back projection reconstruction on the 3D X-RAY data acquired by continuous spiral scanning to obtain initial three-dimensional volume data. The initial three-dimensional volume data is then optimized using an iterative reconstruction algorithm to generate full-plate three-dimensional spatial data.

[0067] The preprocessing module is used to denoise the entire 3D spatial data using adaptive 3D Gaussian filtering. It adjusts the filter kernel size according to the local variance of the voxels to obtain denoised 3D spatial data. The denoised 3D spatial data is then divided into several 3D sub-blocks. Histogram transformation is performed on each 3D sub-block. Gray levels exceeding a set value within each 3D sub-block are clipped, and the clipped portions are redistributed to the remaining gray levels in the histogram of that 3D sub-block. The contrast variation of each 3D sub-block is controlled to obtain transformed 3D spatial data. Finally, interpolation smoothing is performed on the boundaries between the 3D sub-blocks in the transformed 3D spatial data to obtain 3D spatial data with enhanced material contrast.

[0068] The feature extraction module is used to locate the target analysis region of each resin plug pore based on the three-dimensional spatial data with enhanced material contrast. It performs three-scale Gaussian pyramid decomposition on each target analysis region to obtain the internal microstructure features at the original resolution scale, the interface features at a 2x downsampling scale, and the surface morphology features at a 4x downsampling scale. The internal microstructure features, interface features, and surface morphology features are fused according to preset weights. Different fusion weights are set for internal defects and surface defects to obtain the fused features of each target analysis region. Based on the fused features, multidimensional features of each plug pore are extracted. The multidimensional features include surface morphology features, internal structure features, and interface bonding features.

[0069] The defect identification module is used to identify surface defects, internal defects, and interface defects of each plug hole based on multi-dimensional features and quantify the corresponding defect parameters. It also identifies full-board missing plug, half-plug, and plug hole position offset defects based on the comparison between the designed hole position and the actual hole position detected from the three-dimensional spatial data.

[0070] Third Embodiment A method for evaluating the quality of PCB resin via plugs based on the detection method in the first embodiment includes the following steps: Obtain the defect identification results and corresponding defect quantification parameters for each via on the PCB board under test. The defect identification results include whether each via has a defect and the type of defect. The defect quantification parameters include indentation depth, protrusion height, equivalent diameter of bubble, bubble volume, poor bonding area, and the proportion of poor bonding area to the total interface area.

[0071] According to a preset grading standard, defects of various types in each plugged hole are classified into severity levels. In the preset grading standard, defects are divided into fatal defects, severe defects, general defects, and minor defects from high to low severity. Fatal defects include incomplete plugging, partial plugging, and air bubbles exceeding a set size; severe defects include poor bonding area exceeding a set proportion and resin cracking; general defects include air bubbles within a set size range and dents within a set depth range; minor defects include air bubbles smaller than a set size, dents smaller than a set depth, and surface foreign matter. This grading standard provides a unified evaluation basis for subsequent overall board acceptance judgment and process anomaly location.

[0072] Quality was evaluated from three dimensions: board level, single-hole level, and process level. Board-level evaluation: This involves statistically analyzing the quantity, density, and location of all types of defects across the entire board, generating a defect distribution heatmap. The heatmap uses varying shades or different colored blocks to mark the distribution of various defects on the board's coordinate system. Areas with concentrated defects are highlighted, allowing process engineers to visually locate abnormal areas and preliminarily determine the source of the anomaly, such as specific equipment areas or material batch differences.

[0073] Single-hole level evaluation: The depression depth, equivalent diameter of air bubbles, and proportion of poor bonding area for each plugged hole are quantitatively scored. The measured values ​​of each parameter are mapped to a unified scoring scale, and the weighted sum is used to obtain the single-hole quality score for each plugged hole. This score can intuitively compare the filling quality of different plugged holes, providing a convenient reference for subsequent engineering analysis.

[0074] Process-level evaluation: Calculate the process capability index (Cpk) of key parameters for normal via plugging across the entire board to identify areas of process fluctuation. Treat parameters such as the depression depth and bubble size of normal via plugging across the entire board as a set of measurement data, and calculate its Cpk value. This value reflects the stability and consistency of the current process. When the Cpk value is lower than a set threshold, it indicates a systematic deviation in the process, requiring adjustment by tracing the abnormal areas in the thermal map back to the specific process step.

[0075] Based on the results of board-level evaluation, single-hole-level evaluation, and process-level evaluation, a comprehensive quality score for the entire board is calculated according to preset weights to determine whether the PCB board under test is qualified or unqualified. The comprehensive quality score integrates the evaluation results of the three dimensions into a single numerical index, facilitating a quick determination of whether the entire board meets the shipping standards.

[0076] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0077] In addition, the functional modules in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0078] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for detecting PCB resin via plugging defects based on multi-scale 3D feature fusion, characterized in that, The method, applied to a 3D X-ray inspection system including an X-ray source, detector, and motion platform, comprises the following steps: The X-ray source and detector are controlled to rotate synchronously, and the motion platform is controlled to drive the PCB board under test to move at a constant speed. The PCB board under test is continuously spiral scanned to collect the full board 3D X-ray data, which is then reconstructed to generate the full board three-dimensional spatial data. The three-dimensional spatial data is preprocessed with adaptive Gaussian denoising and contrast enhancement to obtain three-dimensional spatial data with enhanced material contrast, and the target analysis area where each plug hole is located is located by grayscale threshold segmentation. Multi-scale Gaussian pyramid decomposition of each target analysis region is performed and then fused with attention weighting to extract multi-dimensional features of each plug, including surface morphology features, internal structure features and copper-resin interface features. Based on the multidimensional features, surface defects, internal defects, and interface defects of each plug hole are identified and the corresponding defect parameters are quantified; and based on the comparison between the designed hole position and the actual hole position detected from the three-dimensional spatial data, defects such as full board leakage plugging, partial plugging, and plug hole position offset are identified.

2. The PCB resin plugging defect detection method according to claim 1, characterized in that, The process of reconstructing and generating full-panel three-dimensional spatial data includes the following steps: The 3D X-RAY data is filtered and back-projected to reconstruct the initial three-dimensional volume data. An iterative reconstruction algorithm is used to optimize the initial three-dimensional volume data to generate full-plate three-dimensional spatial data.

3. The PCB resin plugging defect detection method according to claim 1, characterized in that, The enhanced three-dimensional spatial data for material contrast is obtained through the following steps: An adaptive three-dimensional Gaussian filter is used to denoise the full-plate three-dimensional spatial data. The filter kernel size is adjusted according to the local variance of the voxels to obtain the denoised three-dimensional spatial data. The denoised 3D spatial data is divided into several 3D sub-blocks. Histogram transformation is performed on each 3D sub-block. Gray levels exceeding a set value in each 3D sub-block are clipped. The clipped parts are redistributed to the remaining gray levels in the histogram of each 3D sub-block. The contrast variation of each 3D sub-block is controlled to obtain the transformed 3D spatial data. Interpolation smoothing is performed on the boundaries between the three-dimensional sub-blocks in the transformed three-dimensional spatial data to obtain three-dimensional spatial data with enhanced material contrast.

4. The PCB resin plugging defect detection method according to claim 1, characterized in that, The method of locating the target analysis region where each resin plug pore is located by segmenting and locating the three-dimensional spatial data based on the material contrast enhancement includes the following steps: Based on the grayscale difference between copper foil and resin in X-ray attenuation coefficient, a grayscale threshold is set for the three-dimensional spatial data of the material with enhanced contrast. Regions with grayscale values ​​greater than the threshold are divided into copper foil regions, and regions with grayscale values ​​less than the threshold are divided into resin-filled regions. Based on the geometry of each copper foil region, identify the orifice location and wall extent of each via; Using the center of each through hole as the center, the radius as the set ratio of the hole diameter, and the plate thickness as the height, a three-dimensional sub-region containing the complete through hole structure is extracted from the three-dimensional spatial data as the target analysis area for each resin through hole.

5. The PCB resin plugging defect detection method according to claim 1, characterized in that, The full-plate leakage and partial-plate defects are identified through the following steps: Read the coordinates of the designed holes on the PCB board under test; Based on the grayscale features of the copper holes in the three-dimensional spatial data with enhanced material contrast, the coordinates of all actual hole locations are detected. Perform nearest neighbor matching between the designed hole positions and the actual hole positions; If the designed hole position does not match the actual hole position, it is judged as a leakage plug defect; If the designed hole position matches the actual hole position, but the resin filling height in the actual hole position is less than the set percentage of the plate thickness, it is determined to be a partial plug defect.

6. The PCB resin plugging defect detection method according to claim 5, characterized in that, The plug hole position offset defect is identified through the following steps: For successfully matched holes, the actual hole coordinates are compared with the corresponding designed hole coordinates; If the actual hole position coordinates are offset from the designed hole position coordinates by more than a set threshold, it is determined to be a hole position offset defect.

7. The PCB resin plugging defect detection method according to claim 4, characterized in that, The process of performing multi-scale Gaussian pyramid decomposition on each target analysis region and then performing attention-weighted fusion to extract multidimensional features of each resin plug pore includes: Three-scale Gaussian pyramid decomposition was performed on each target analysis region to obtain the internal microstructure features at the original resolution scale, the interface features at a 2x downsampling scale, and the surface morphology features at a 4x downsampling scale. The internal microstructure features, interface features and surface morphology features are fused according to preset weights, wherein different fusion weights are set for internal defects and surface defects respectively, to obtain the fused features of each target analysis region. Based on the fusion features, multidimensional features of each plug hole are extracted, including surface morphology features, internal structure features, and interface bonding features. The surface morphology features include the height distribution of the plugged hole surface and the relative height parameter based on the reference plane; the internal structural features include microbubble regions, linear edges of microcracks, and grayscale anomalies within the resin region; and the interface bonding features include the grayscale gradient distribution and gradient anomaly regions at the interface between the copper foil and the resin.

8. The PCB resin plugging defect detection method according to claim 7, characterized in that, The methods for identifying and quantifying surface defects include: Using the direction perpendicular to the PCB board surface as the Z-axis, the Z-axis coordinates of the resin surface in each plane position within the via filling area are obtained layer by layer to obtain the height distribution data of the via filling surface. The height data of the copper foil area around the orifice is fitted with a plane to obtain the reference plane; Calculate the height difference between the resin surface and the reference surface to obtain the depression depth or protrusion height.

9. The PCB resin plugging defect detection method according to claim 7, characterized in that, The methods for identifying and quantifying internal defects include: Using the direction perpendicular to the PCB board surface as the Z-axis, extract multiple Z-axis slices with equal spacing along the hole axis with a set step size; Resin regions in each slice are segmented based on grayscale thresholding; A three-dimensional region growth algorithm is used to grow connected regions within the resin region where the gray value is lower than a set threshold, thereby obtaining the three-dimensional region of the bubble and calculating its equivalent diameter. A three-dimensional edge detection algorithm is used to extract the linear edge structure within the resin area to identify resin cracks.

10. The PCB resin plugging defect detection method according to claim 7, characterized in that, The methods for identifying and quantifying interface defects include: Based on the grayscale gradient of the three-dimensional spatial data with enhanced material contrast, the interface between the copper foil and the resin is extracted. Calculate the grayscale gradient value at each location on the interface; When the grayscale gradient value of a certain area is lower than the set threshold, the area is determined to be a poorly bound area; Calculate the area of ​​the poorly bonded region and its proportion of the total interface area.

11. A PCB resin plugging defect detection system for implementing the method according to any one of claims 1-10, characterized in that, include: An X-ray source used to emit X-rays; A detector is used to receive X-rays that penetrate the PCB board under test and convert them into electrical signals. The motion platform is used to support the PCB board under test and drive it to move at a constant speed during continuous spiral scanning. The GPU parallel data processing unit is used to filter and back-project the 3D X-RAY data acquired by continuous spiral scanning, and combine iterative reconstruction to generate full-board three-dimensional spatial data. It is also used to perform adaptive Gaussian denoising and contrast enhancement preprocessing, multi-scale Gaussian pyramid decomposition and attention-weighted fusion, as well as defect identification and quantization on the three-dimensional spatial data.

12. A method for evaluating the quality of PCB resin via plugging based on the detection method according to any one of claims 1-10, characterized in that, The steps include the following: Obtain the defect identification results and corresponding defect quantification parameters for each plug hole on the PCB board under test; According to the preset grading standard, the severity level of each type of defect in each plug hole is classified. Quality is evaluated from three dimensions: board level, single hole level, and process level. The overall quality score of the board is calculated based on the evaluation results to determine whether the PCB board under test is qualified or unqualified. In the preset grading standard, defects are divided into fatal defects, serious defects, general defects and minor defects from high to low. Fatal defects include leaks, partial blockages and bubbles exceeding the set size. Serious defects include poor bonding area exceeding the set ratio and resin cracking; Common defects include bubbles within a specified size range and dents within a specified depth range; Minor defects include bubbles smaller than the set size, dents less than the set depth, and surface foreign objects.

Images