A multi-parameter detection and analysis method for a circuit board vacuum film pressing process

Abstract

The application discloses a kind of line board vacuum film pressing process multi-parameter detection analysis method, it is related to line board vacuum film pressing process parameter detection field, comprising: the present application is by main controller synchronous acquisition instruction at the start moment of film pressing process, control PLC reads vacuum degree, pressure and temperature data simultaneously with same scanning period, and add uniform time stamp for each reading data;Subdivision gridding processing is carried out to the optical image on the substrate surface, and potential defect area is identified based on neighborhood window gray scale variance, and the connected domain profile of bubble and wrinkle is distinguished by combining morphological open operation;By extracting the gradient of uniformity characteristic value of dry film to identify the deterioration diffusion path, and analyze the risk type of abnormal defect, and combined with process parameter time series data, the process parameters leading to quality risk are determined by correlation analysis, multiple parameters are synchronously collected, defect identification and process quality dynamic correlation are realized, and the quality risk root can be accurately traced.

Description

Technical Field

[0001] This invention belongs to the field of circuit board vacuum lamination process parameter detection technology, and relates to a multi-parameter detection and analysis method for circuit board vacuum lamination process. Background Technology

[0002] In the printed circuit board manufacturing process, the vacuum lamination process uses heating, pressurization, and vacuuming to uniformly adhere a photosensitive dry film to the substrate surface, forming the mask layer required for subsequent pattern transfer. The quality of this process directly determines the precision and reliability of circuit pattern etching or electroplating.

[0003] Currently, the following technical bottlenecks exist in the detection and analysis technology for vacuum lamination processes of printed circuit boards (PCBs): First, in actual production, the key process parameters of PCB vacuum lamination are in a continuous and dynamic coupling change. Existing technologies typically lack an effective multi-parameter synchronous acquisition mechanism, resulting in data from different sources not being accurately aligned in time sequence, making it difficult to accurately capture the co-evolution trend between parameters. Therefore, when process drift or intermittent disturbances occur, it is difficult to trace the root cause of the anomaly, easily leading to incorrect quality attribution, thus lacking direction for process optimization.

[0004] Secondly, when identifying surface defects based on images, existing technologies often rely on global thresholds, treating pixel grayscale information as an isolated condition and ignoring the dynamic changes in background noise caused by uneven lighting, material differences, and inherent textures. This approach easily misjudges occasional grayscale fluctuations as defects, fails to reliably distinguish between real defects and background interference, resulting in poor detection reliability and potential quality risks due to missed detections.

[0005] Finally, existing technologies typically only count and record the number and size of individual defects in isolation, lacking analysis of the spatial distribution patterns of defects. This makes it impossible to systematically identify the clustering trend of defects caused by equipment or process abnormalities, and it is also difficult to provide early warnings of potential risks such as uniform degradation. When quality control is severely lagging, it often has to rely on post-event remediation, making it difficult to achieve pre-event prevention and precise intervention, thus restricting production efficiency. Summary of the Invention

[0006] In view of this, in order to solve the problems mentioned in the background art, the present invention provides a multi-parameter detection and analysis method for circuit board vacuum lamination process.

[0007] The objective of this invention can be achieved through the following technical solution: a multi-parameter detection and analysis method for vacuum lamination process of circuit boards, comprising: S1, during the vacuum lamination process, synchronously collecting time-series data of process parameters such as vacuum degree of lamination cycle, lamination pressure and lamination temperature.

[0008] S2. After lamination is completed, an optical image of the substrate surface after lamination is acquired using a linear scan camera.

[0009] S3. Perform partitioned meshing processing on the optical image of the substrate surface, extract the dry film uniformity feature value, identify the connected domain contours of the bubble and wrinkle regions in the optical image of the substrate surface, and calculate the abnormal morphological parameters of the connected domain contours.

[0010] S4. Calculate the gradient of dry film uniformity characteristics between adjacent grids in the optical image of the substrate surface to identify the gradient and direction of uniformity degradation diffusion, and locate the area of ​​dry film adhesion uniformity degradation accordingly.

[0011] S5. Analyze the spatial distribution density of abnormal morphological characteristics to determine the type of abnormal defect risk.

[0012] S6. Based on the abnormal defect risk type and the deterioration area of ​​dry film adhesion uniformity, and combined with the time series data of process parameters, determine the process parameters that lead to quality risks through correlation analysis.

[0013] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) The present invention controls the PLC to read vacuum, pressure and temperature data at the same scan cycle by issuing a synchronous acquisition command from the main controller at the beginning of the pressing process, and adds a unified timestamp to each data read, thereby aligning multiple process parameters on the time axis, capturing the collaborative evolution trend between parameters, and retaining the causal time sequence information of the process state. Thus, when the process drifts or intermittent disturbances occur, the root cause of the abnormality can be traced, providing direction for process optimization and effectively avoiding quality attribution errors.

[0014] (2) The present invention performs partitioned meshing processing on the optical image of the substrate surface, identifies potential defect areas based on the gray variance of the neighborhood window, and distinguishes the connected region contours of bubbles and wrinkles by combining morphological opening operation. It dynamically adapts to the accidental gray fluctuations caused by uneven illumination and material differences, accurately identifies physical defects such as bubbles and wrinkles, and effectively reduces the risk of false detection.

[0015] (3) This invention extracts the gradient of dry film uniformity feature value to identify the degradation diffusion path, analyzes the abnormal defect risk type, and combines the time series data of process parameters to determine the process parameters that lead to quality risks through correlation analysis. This enables the systematic identification of defect aggregation trends and early warning of uniform degradation, and establishes a dynamic correlation between process parameters and quality risks. This shifts quality control from post-event remediation to pre-event prevention, effectively improving production efficiency. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a diagram illustrating the implementation steps of the method of the present invention.

[0018] Figure 2 This is a flowchart for determining the diffusion direction of the dry film uniformity distribution.

[0019] Figure 3 A flowchart for defining the deterioration region of dry film adhesion uniformity in this invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Please see Figure 1 As shown, the present invention provides a multi-parameter detection and analysis method for vacuum lamination process of circuit board, including: S1, during the vacuum lamination process, synchronously collecting the timing data of process parameters such as vacuum degree of lamination cycle, lamination pressure and lamination temperature.

[0022] To ensure the temporal consistency of subsequent correlation analysis and to establish the causal relationship between process parameters and quality defects, in a specific embodiment of the present invention, the acquisition of process parameter time-series data is achieved through the following synchronization and time-series alignment mechanism: at the start of the vacuum molding process, the main controller of the molding equipment issues a synchronization acquisition command.

[0023] In response to the synchronous acquisition command, the programmable logic controller inside the device simultaneously reads real-time measurement values ​​from the vacuum gauge, pressure sensor, and temperature sensor at the same scan cycle.

[0024] The vacuum gauge is installed on the vacuum pipeline connected to the chamber to monitor the vacuum level in the chamber in real time; the pressure sensor is integrated into the pressure plate of the pressing mechanism to directly detect the pressing pressure applied to the substrate surface; the temperature sensor is set at a temperature measuring point close to the substrate surface to monitor the temperature change during the pressing process.

[0025] The programmable logic controller adds a unified timestamp to each synchronously read vacuum, pressure, and temperature data, forming a timestamped raw data point.

[0026] To achieve synchronous acquisition of multiple parameters, the programmable logic controller (PLC) is configured with a high-speed, cyclically executing data acquisition task and a timed interrupt program. Upon receiving a synchronous acquisition command, the PLC triggers an interrupt at the beginning of each scan cycle, reads the latest cached data from all configured sensor channels in parallel, and immediately adds the same timestamp to this set of data. This ensures strict alignment of vacuum, pressure, and temperature data on the time axis, providing an accurate data foundation for subsequent time-series correlation analysis.

[0027] The unified timestamp originates from the PLC's internal real-time clock or is provided by an external high-precision time server connected to it. After each synchronous reading of a set of vacuum, pressure, and temperature data, the PLC immediately calls the clock function block to obtain the current absolute time and packages the timestamp and the three measured values ​​into a structured data record, which is then stored in a designated data buffer. This ensures that each process parameter data has a consistent time base, providing time coordinates for subsequent time series correlation analysis.

[0028] Continue performing the above steps until the molding process is complete, generating a timeline curve of process parameters with three time axes aligned.

[0029] S2. After lamination is completed, an optical image of the substrate surface after lamination is acquired using a linear scan camera.

[0030] To avoid image loss or redundancy due to improper acquisition timing and to ensure the reliability of subsequent image stitching and analysis, the steps for acquiring optical images of the substrate surface after lamination are as follows: The laminated substrate is controlled to be transported at a constant speed to the image acquisition station. When the photoelectric position sensor at the station entrance detects the leading edge position of the substrate, the strip light source is activated to illuminate the substrate surface and drive the linear scan camera to continuously acquire surface image sequences.

[0031] The image acquisition in this invention is performed at a separate image acquisition station, which integrates a conveyor belt, a trigger sensor, an illumination unit, and an image acquisition unit. A photoelectric position sensor is installed at the station entrance, with its optical path spanning the conveyor belt path, used to detect the arrival time of the substrate's leading edge; the illumination direction of the strip light source unit covers the substrate surface area above the conveyor device; the optical axis of the linear scan camera unit is perpendicular to the substrate surface, and its scanning line direction is perpendicular to the substrate conveying direction, ensuring that the substrate can be scanned sequentially when it enters.

[0032] It should be noted that the uniform speed of transmission to the image acquisition station is to ensure that the line frequency of the linear scan camera and the speed of the substrate movement remain in a constant correspondence, thereby avoiding geometric distortion of the acquired image sequence during stitching. For example, the speed can be set in the range of 100 to 1000 mm / s.

[0033] The substrate surface is illuminated by a strip light source because it provides a uniform and stable light field along the scanning line, which helps the line scan camera obtain a consistent image grayscale response during continuous scanning, thereby improving the reliability of subsequent image analysis and feature extraction.

[0034] When the end of the substrate has completely passed the image acquisition station, the linear scan camera stops acquiring images.

[0035] During the acquisition process, the acquisition should end precisely when the end of the substrate has completely passed the image acquisition station. To achieve this, for example, a control logic based on timing calculation can be used: a timer in the PLC is started synchronously at the moment the photoelectric position sensor triggers the start of acquisition; based on the substrate length and constant conveying speed, the theoretical time for the substrate to completely pass through is calculated; and a stop command is issued to the camera when this time expires. As another preferred embodiment, to improve adaptability to substrates of different specifications and achieve boundary control, a second photoelectric sensor can be installed at the exit of the image acquisition station. When this photoelectric sensor detects that the end of the substrate has left, a stop trigger signal is immediately generated.

[0036] The surface image sequence is stitched together in the order of acquisition to generate a two-dimensional image of the substrate surface.

[0037] The stitching process is as follows: First, the ratio of the constant substrate transport speed to the fixed line frequency of the linear scan camera is calculated, which serves as the physical spacing between adjacent line scan images in the transport direction. Then, each row of image data in the sequence is mapped to the corresponding Y-axis coordinate in the two-dimensional image matrix according to its acquisition order. n represents the sequence number, n=0,1,2... The value represents the Y-axis coordinate of the nth row of pixel data in the final generated 2D panoramic image, which determines the vertical position of this row of data in the final composite image. Then, the transformation from temporal data to spatial image is completed by overlaying pixel rows, thereby generating a coordinate-continuous 2D panoramic image.

[0038] S3. Perform partitioned meshing processing on the optical image of the substrate surface, extract the dry film uniformity feature value, identify the connected domain contours of the bubble and wrinkle regions in the optical image of the substrate surface, and calculate the abnormal morphological parameters of the connected domain contours.

[0039] To achieve spatial quantitative characterization of the uniformity of dry film adhesion on the substrate surface, capture the local differences and overall patterns of dry film distribution, and provide standardized quantitative data support for subsequent analysis of the diffusion gradient and direction of dry film adhesion uniformity degradation and the correlation between quality risks and process parameters, the steps for extracting dry film uniformity feature values ​​are as follows: Based on the preset number of grid rows and columns, the two-dimensional image of the substrate surface is divided into multiple uniform rectangular grid arrays, where each rectangular grid region corresponds to an actual physical location region on the substrate surface.

[0040] It should be noted that the preset number of grid rows and columns can be determined based on the nominal size of the substrate and the minimum analytical unit size required for process quality control. For example, if the detection system is required to distinguish the uniformity variation in a 1mm × 1mm area, and the substrate size is 300mm × 400mm, then the preset number of grid rows can be 400 and the number of columns can be 300.

[0041] Calculate the arithmetic mean of grayscale values ​​of all pixels within the grid region of each rectangular grid array, and use this arithmetic mean as the dry film uniformity feature value of the current grid.

[0042] A double loop structure can be used to traverse the pixels within a grid and calculate the average gray value. This involves visiting each grid in turn, traversing the pixels within the grid, accumulating the gray values, and then dividing by the total number of pixels.

[0043] It should be added that, under standardized coaxial or diffuse lighting conditions, the absorption and scattering effect of the dry film on the incident light in its local area, as a semi-transparent medium, determines the light intensity reflected to the camera. The arithmetic mean of the gray values ​​of the image in that area reflects the average optical density of the dry film in that area. The optical density is positively correlated with the physical thickness of the dry film and the density of its material. Therefore, this average value can characterize the overall coverage of the dry film in that area.

[0044] Based on the spatial arrangement order of the rectangular grid array from left to right and from top to bottom in the two-dimensional image of the substrate surface, the dry film uniformity characteristic values ​​of each grid are arranged sequentially to form a two-dimensional array.

[0045] The rows and columns of this two-dimensional array correspond to the spatial row and column distribution of the grid, forming a dry film uniformity characteristic distribution matrix.

[0046] To identify two types of defects—bubbles and wrinkles—formed during the lamination process on the substrate surface, this study aims to filter out image noise that interferes with defect detection and clearly distinguish defect type boundaries, providing accurate and effective defect area data support for subsequent calculation of abnormal morphological parameters and determination of defect risk types.

[0047] Based on this, the steps for identifying the connected region contours of the bubble and wrinkle areas are as follows: preprocess the optical image of the substrate surface to obtain an enhanced grayscale image.

[0048] The purpose of image preprocessing is to suppress noise and enhance the contrast between defect areas and the background, thereby standardizing image quality. Specifically, this includes: first, performing median filtering on the original optical image to eliminate noise that may be introduced during imaging; then, using histogram equalization to enhance the overall or local grayscale dynamic range of the image, thereby improving the distinguishability of defect features such as bubbles and wrinkles from a uniform background; finally, the output is a grayscale image that has undergone noise reduction and contrast enhancement, used for subsequent texture analysis and defect segmentation.

[0049] Define a neighborhood window centered on each pixel in the enhanced grayscale image, and calculate the variance of grayscale values ​​of all pixels within the neighborhood window.

[0050] The neighborhood window is a k×k pixel rectangular window centered on the current pixel, where k is an odd number greater than 1, and its size is set according to the texture roughness of the defect to be detected.

[0051] If the grayscale variance is greater than the texture variance threshold, the pixel is determined to be located in a potential defect area of ​​texture abrupt change and is marked as a foreground point; otherwise, it is marked as a background point. This generates a preliminary first binary image of the bubble and wrinkle abnormal areas, which can initially identify texture abrupt change areas such as local bright bulges caused by bubbles and dark depressions caused by wrinkles from a uniform background.

[0052] It should be added that the texture variance threshold is based on the statistical analysis of the gray-level variance of the 3×3 neighborhood windows in 200 normal film-pressed sample images, and the 99th percentile of the variance value is taken as the threshold.

[0053] A morphological opening operation is performed on the first binary image to eliminate noise and smooth region boundaries, resulting in a second binary image. Connected regions formed by all foreground points are then searched in the second binary image as candidate defect regions.

[0054] In the first and second values ​​obtained from the initial segmentation, in addition to the real bubble or wrinkle areas, there are often several isolated foreground points and irregular area boundaries due to image noise, uneven lighting or texture misjudgment. These factors can interfere with the judgment of the actual number, location and shape of defects.

[0055] The morphological opening operation consists of a sequence of operations: erosion followed by dilation. The erosion operation uses a structuring element of shape and size to convolve the first binary image, traversing the image to eliminate isolated foreground points smaller than the structuring element, thereby filtering out noise. Subsequently, the dilation operation uses the same structuring element to traverse the eroded image, restoring the size of the real defect areas that were over-eroded and making their boundaries smoother and more continuous, ultimately obtaining the second binary image.

[0056] For each candidate defect region, the corresponding local region is located by tracing back to the enhanced grayscale image, and the grayscale mean of the local region is calculated. If the grayscale mean is higher than the high grayscale threshold, it is determined to be a bubble region; if the grayscale mean is lower than the low grayscale threshold, it is determined to be a wrinkle region.

[0057] In optical images, bubble areas typically exhibit high brightness due to enhanced specular reflection caused by their protrusions, with a significantly higher average grayscale value than normal areas; wrinkled areas typically exhibit low brightness due to shadows caused by their depressions or accumulations.

[0058] The high and low grayscale thresholds can be determined through statistical analysis of known defective samples. For example, the mean and standard deviation of grayscale values ​​in a large number of normal sample areas can be calculated. The mean grayscale value of the normal area plus twice the standard deviation can be set as the lower limit of the high threshold, and the mean grayscale value minus twice the standard deviation can be set as the upper limit of the low threshold. If the grayscale value falls between the low and high grayscale thresholds, it is classified as background.

[0059] A specific embodiment of the present invention for calculating the abnormal morphological parameters of a connected region contour includes: calculating the coordinates of the center point of the connected region contour as the location parameter of the defect.

[0060] The area of ​​the connected region's outline is obtained by counting all the pixels enclosed within the outline and multiplying that number by the actual area represented by a single pixel.

[0061] Calculate the ratio of the length of the longer side to the shorter side of the smallest bounding rectangle of the connected region outline, and use it as its aspect ratio.

[0062] S4. Calculate the gradient of dry film uniformity characteristics between adjacent grids in the optical image of the substrate surface to identify the gradient and direction of uniformity degradation diffusion, and locate the area of ​​dry film adhesion uniformity degradation accordingly.

[0063] See Figure 2 As shown, given the propagation intensity and spatial evolution of dry film uniformity degradation, clarifying the gradient change trend and diffusion direction of degradation characteristics, eliminating the ambiguity of degradation diffusion mode, and identifying the gradient and direction of uniformity degradation diffusion include: for each grid cell in the dry film uniformity characteristic distribution matrix, calculating the difference of dry film uniformity characteristic values ​​with adjacent grid cells along the row and column directions of the matrix, forming a horizontal gradient matrix and a vertical gradient matrix.

[0064] The difference in the eigenvalues ​​of dry film uniformity represents the intensity and trend of local variation in dry film uniformity in the corresponding direction; positive values ​​indicate increasing eigenvalues, and negative values ​​indicate decreasing eigenvalues. The calculated matrices together constitute a two-dimensional gradient field representing the spatial rate of change of uniformity.

[0065] Traverse the horizontal and vertical gradient matrices and extract the grid sequence with the same gradient value sign and continuous spatial position as the local change vector.

[0066] The same gradient value sign means that the positive or negative sign of the gradient value remains consistent between adjacent grid cells along the row or column direction of the matrix; spatial continuity means that the grid cells corresponding to these gradient values ​​with the same sign share a common edge or corner in the two-dimensional spatial array. This distinguishes isolated gradient points from physical processes that reflect the continuous enhancement or weakening of dry film uniformity characteristics along a certain direction.

[0067] Connecting and merging spatially adjacent local change vectors with the same gradient sign forms a continuous path that reflects the monotonic change of eigenvalues, serving as the change propagation path.

[0068] The rules for determining connection and merging are as follows: if two local change vectors have at least one pair of directly adjacent grid cells in space, that is, the cells share a common edge and the common difference sign of the two is the same, then the two can be merged, and the grid coordinate sequences of the two vectors that meet the conditions are spliced ​​together to form an ordered coordinate sequence, and their common difference sign is inherited.

[0069] If all signs of the local change vectors in the change propagation path are positive, it is determined to be positive diffusion; if all signs are negative, it is determined to be negative diffusion. The principal component direction of the set of grid cells covered by the change propagation path is defined as the main diffusion direction.

[0070] The difference values ​​of all local change vectors in the path have the same sign, reflecting the spatial variation trend of the dry film uniformity characteristic value along the path. If all signs are positive, it indicates that the characteristic value increases monotonically along the path, corresponding to the positive diffusion of gradual optimization of dry film coverage quality; if all signs are negative, it indicates that the characteristic value decreases monotonically, corresponding to the negative diffusion of gradual deterioration of dry film quality.

[0071] The main diffusion direction is determined by calculating the principal components of the spatial coordinates of the set of grid cells covered by the path. Specifically, this includes: constructing a point set from the coordinates of the center points of the cells, calculating its covariance matrix, and performing eigenvalue decomposition on the matrix; taking the direction of the eigenvector corresponding to the largest eigenvalue, which is the first principal component direction of the point set distribution, i.e., the main diffusion direction. This direction spatially represents the overall extension trend of the change propagation path.

[0072] See Figure 3As shown, due to the lack of a clear definition standard for the spatial coverage of dry film uniformity degradation, it is easy to miss degraded areas or misjudge non-degraded areas, which in turn leads to deviations in subsequent quality risk type determination and defocusing of process parameter correlation analysis, seriously affecting the reliability of detection and analysis results. Therefore, the steps for locating the degraded areas of dry film adhesion uniformity include: delineating the grid cells where the dry film uniformity characteristic values ​​of each change propagation path show a monotonically changing trend, and the grid cells that are directly adjacent to any grid cell in the path and whose dry film uniformity characteristic values ​​are within the range of characteristic values ​​at both ends of the path, as the preliminary affected areas.

[0073] The diffusion of dry film uniformity degradation follows the gradient transfer law. Grid cells whose dry film uniformity characteristic values ​​show a monotonically changing trend along the propagation path are the core areas affected by degradation diffusion. The continuous monotonous change of their characteristic values ​​reflects the propagation trajectory and intensity of degradation. Grid cells that are directly adjacent to any grid cell in this path and whose dry film uniformity characteristic values ​​are within the range of characteristic values ​​at both ends of the path are in the transition zone of the degradation gradient. Their characteristic values ​​do not undergo abrupt changes and have a continuous gradient correlation with the core degradation area. This indicates that these grid cells have been indirectly affected by the degradation diffusion and belong to the potential area for degradation extension. Delineating the above two types of grid cells together as the initial affected area not only conforms to the spatial propagation physical characteristics of dry film adhesion uniformity degradation, but also ensures the completeness of capturing the degradation impact range and avoids the narrow definition of the degradation area due to the omission of the transition area.

[0074] If any two regions in the initial affected area have adjacent or overlapping grid cells, these two regions will be merged to form a merged region, which will be identified as a region of deteriorated dry film adhesion uniformity.

[0075] In vacuum lamination processes, the deterioration of dry film adhesion caused by the same process parameter anomaly typically manifests as a continuous and complete physical region in space. If two initially affected regions are spatially adjacent or overlap, it indicates that they are physically continuous and likely originate from the same or closely related source of the anomaly. Therefore, merging them into a single region of dry film adhesion uniformity deterioration based on the principle of spatial continuity can fully represent the actual range of deterioration caused by a specific process anomaly.

[0076] S5. Analyze the spatial distribution density of abnormal morphological characteristics to determine the type of abnormal defect risk.

[0077] Specifically, the average nearest neighbor distance of all defect center points is calculated. If the average nearest neighbor distance is less than the expected nearest neighbor distance, it is determined to be a spatial clustering pattern; otherwise, it is determined to be a spatial dispersion pattern.

[0078] The average nearest neighbor distance is calculated by taking the minimum Euclidean distance between the center point of each defect and the center points of all other defects, and taking that minimum as the nearest neighbor distance for that point.

[0079] The expected nearest neighbor distance refers to the theoretically predicted average nearest neighbor distance when an equal number of points fall within the same region under ideal conditions of complete randomness, i.e., following a spatial Poisson distribution. This value serves as a benchmark for random distribution comparison with actual observations. Its calculation formula, based on spatial statistics theory, is as follows: Where ρ is the surface density of defects, i.e., the ratio of the total number of defects N to the effective analytical area A of the substrate. Therefore, the formula can also be directly expressed as: In the formula, area A is the actual physical area of ​​the analyzed region on the substrate surface, and D... exp This represents the expected average distance between points under completely random conditions with no tendency to cluster or disperse.

[0080] If the average nearest neighbor distance is less than the expected nearest neighbor distance, it indicates that the average distance between actual defect points is significantly less than the expected distance under a completely random distribution, indicating that the defect points tend to move closer to each other in space. Therefore, the distribution of defects is determined to be a spatial clustering pattern.

[0081] Conversely, it indicates that the average distance between actual defect points is equal to or greater than the expected distance of random distribution, and there is no obvious mutual attraction trend between defect points, so it is judged as a spatial dispersion pattern.

[0082] Calculate the ratio of the average absolute deviation of the defect area to its arithmetic mean, and use it as the area dispersion. Similarly, calculate the aspect ratio dispersion.

[0083] If the area dispersion is less than the aspect ratio dispersion, it is judged as a high similarity pattern; otherwise, it is judged as a low similarity pattern.

[0084] Area dispersion indicates the consistency of defects on a physical scale, while aspect ratio dispersion indicates the consistency of geometric shape. Systematic process deviations usually tend to produce defects of similar size, while the shape may fluctuate due to the influence of local microenvironment. Therefore, when area dispersion is less than aspect ratio dispersion, it indicates that the scale similarity of the defect group is higher than the shape similarity, which is more consistent with the characteristics of a single systematic cause, and is thus judged as a high similarity pattern. Conversely, it indicates that the cause of the defects is more complex or random, and is thus judged as a low similarity pattern.

[0085] If the spatial distribution of defects is dispersed and the morphological parameters are of low similarity, then it is determined to be a random and occasional risk type.

[0086] The dispersed spatial distribution pattern indicates that there is no spatial correlation between defects, ruling out localized persistent process anomalies; while the low similarity morphological pattern indicates that there are no regular morphological features among defects, ruling out systematic process deviations. Both of these factors together serve as necessary and sufficient conditions, pointing to multiple instantaneous random disturbances as the cause, thus classifying it as a random and accidental risk.

[0087] If the spatial distribution of defects is clustered and the morphological parameters are highly similar, then it is determined to be a systemic or localized risk type.

[0088] The clustered spatial distribution pattern indicates that the defects are caused by local factors related to spatial location, while the highly similar morphological pattern indicates that the defects have consistent physical characteristics. Both point to the defects originating from reproducible process parameter anomalies, thus classifying them as systemic and localized risk types.

[0089] S6. Based on the abnormal defect risk type and the deterioration area of ​​dry film adhesion uniformity, and combined with the time series data of process parameters, determine the process parameters that lead to quality risks through correlation analysis.

[0090] Specifically, based on the type of abnormal defect risk and the spatial location of the deteriorated area of ​​dry film adhesion uniformity, the time interval corresponding to the deteriorated area during the lamination process is determined, and the time sequence data of the process parameters within the time interval is extracted.

[0091] Based on the substrate transport speed *v*, the scanning frequency of the linear scan camera, and the synchronization timestamp of the laminating equipment, a mapping model between the spatial coordinates of the substrate surface and the laminating time is established. First, the spatial boundary coordinates of the deteriorated area of ​​dry film adhesion uniformity in the two-dimensional image of the substrate surface are extracted. For example, using a rectangular coordinate system with the leading edge of the substrate as the x-axis starting point and the left edge as the y-axis starting point, combined with the substrate transport direction, the spatial x-coordinates corresponding to the leading edge and trailing edge of the deteriorated area are calculated. This is then expressed using the formula... , The start time t1 of the deteriorated area entering the molding station and the end time t2 of leaving the molding station are calculated respectively. Here, t0 is the absolute timestamp of the formal start of the molding process, and the time interval [t1, t2] is initially defined.

[0092] For systemic and localized risks, considering the cumulative effect of process parameters, the time interval is extended to [t1-Δt, t2+Δt], where Δt can be 0.5–2s and can be adaptively adjusted according to the film pressing cycle to cover the complete process stage of deterioration. For random and sporadic risks, the single-frame scanning cycle T of the linear scan camera is used as the basis. s The value can be taken as 500–2000 μs, and the time interval is precisely defined as [t1–T]. s ,t1+T s This ensures the capture of instantaneous, pulse-like abnormal parameters.

[0093] Process parameters that continuously deviate from the normal baseline range in the extracted process parameter time series data are identified as abnormal process parameters.

[0094] In steady-state processes, process parameters that continuously deviate from the baseline indicate that the process state has undergone a substantial change. This change has a cumulative effect and causal significance, and is strongly correlated with spatial and systematic quality defects appearing on the substrate. Therefore, it is selected as an abnormal process parameter to be analyzed.

[0095] The normal baseline range for process parameters is calculated based on statistical analysis of time-series data corresponding to 100 consecutive historical normal production batches for parameters such as vacuum, pressure, and temperature, including calculations of mean and standard deviation. Continuous deviation refers to an abnormal state where the process parameter value exceeds its normal baseline range, such as within two data acquisition cycles.

[0096] If the risk is systemic or localized, the parameter with the highest percentage of abnormal state duration within the corresponding time interval is selected as the main related parameter.

[0097] The percentage of abnormal state duration within the corresponding time interval of abnormal process parameters reflects the relative severity of the deviation of the parameter from the normal state during the defect occurrence period. It has the strongest temporal correspondence with the spatial clustering characteristics of defects and is therefore identified as the main associated parameter.

[0098] If the risk is random and occasional, then the parameters that exhibit intermittent pulses within the corresponding time interval of the abnormal process parameters are identified as the main related parameters.

[0099] The intermittent pulse refers to a transient abnormal event in which the parameter value suddenly exceeds its normal reference range within a very short period of time, such as no more than one or several data acquisition cycles, and then returns to the normal range.

[0100] In scenarios involving random and sporadic risks, defects are typically caused by brief, transient disturbances during the production process. Abnormal process parameters exhibiting intermittent pulse characteristics within a corresponding time interval show that the timing of their pulses coincides temporally with the transient mechanisms of random defect generation, and the suddenness of the pulses also matches the spatial dispersion and low morphological similarity of defects. Therefore, by identifying and screening parameters with such intermittent pulse characteristics, they can be determined as key relevant parameters.

[0101] The key related parameters were identified as the process parameters that caused this quality risk.

[0102] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0103] Those skilled in the art will recognize that the modules 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.

[0104] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0105] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0106] Finally, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for multi-parameter detection and analysis of vacuum lamination process for circuit boards, characterized in that: include: S1. During the vacuum pressing process, the timing data of process parameters such as vacuum degree, pressing pressure and pressing temperature during the pressing cycle are collected simultaneously. S2. After lamination is completed, an optical image of the substrate surface after lamination is acquired using a linear scan camera; S3. Perform partitioned meshing processing on the optical image of the substrate surface, extract the dry film uniformity feature value, identify the connected region contour of the bubble and wrinkle regions in the optical image of the substrate surface, and calculate the abnormal morphological parameters of the connected region contour. S4. Calculate the gradient of dry film uniformity characteristic values ​​between adjacent grids of the optical image on the substrate surface, and identify the gradient and direction of uniformity degradation diffusion, thereby locating the area of ​​dry film adhesion uniformity degradation. S5. Analyze the spatial distribution density of abnormal morphological characteristics to determine the type of abnormal defect risk; S6. Based on the abnormal defect risk type and the deterioration area of ​​dry film adhesion uniformity, and combined with the time series data of process parameters, determine the process parameters that lead to quality risks through correlation analysis.

2. The method for multi-parameter detection and analysis of circuit board vacuum lamination process according to claim 1, characterized in that: S1 specifically includes: At the start of the vacuum film pressing process, the main controller of the film pressing equipment issues a synchronous acquisition command; In response to the synchronous acquisition command, the programmable logic controller inside the device simultaneously reads real-time measurement values ​​from the vacuum gauge, pressure sensor, and temperature sensor at the same scan cycle. The programmable logic controller adds a unified timestamp to each synchronously read vacuum, pressure, and temperature data, forming a raw data point with a timestamp. Continue performing the above steps until the molding process is complete, generating a timeline curve of process parameters with three time axes aligned.

3. The multi-parameter detection and analysis method for circuit board vacuum lamination process according to claim 1, characterized in that: The steps for acquiring optical images of the substrate surface after lamination are as follows: The substrate that has completed the lamination process is conveyed to the image acquisition station at a constant speed. When the photoelectric position sensor at the station entrance detects the leading edge position of the substrate, the strip light source is activated to illuminate the substrate surface and drive the linear scan camera to continuously acquire surface image sequences. When the end of the substrate has completely passed the image acquisition station, the linear scan camera stops acquiring images. The surface image sequence is stitched together in the order of acquisition to generate a two-dimensional image of the substrate surface.

4. The method for multi-parameter detection and analysis of circuit board vacuum lamination process according to claim 1, characterized in that: The steps for extracting the uniformity characteristic values ​​of the dry film are as follows: Based on the preset number of rows and columns, the two-dimensional image of the substrate surface is divided into multiple uniform rectangular grid arrays; Calculate the arithmetic mean of gray values ​​of all pixels within the grid region in each rectangular grid array, and use this arithmetic mean of gray values ​​as the dry film uniformity feature value of the current grid. According to the spatial arrangement order of the rectangular grid array from left to right and from top to bottom in the two-dimensional image of the substrate surface, the dry film uniformity characteristic values ​​of each grid are arranged in sequence to form a two-dimensional array; The rows and columns of this two-dimensional array correspond to the spatial row and column distribution of the grid, forming a dry film uniformity characteristic distribution matrix.

5. The multi-parameter detection and analysis method for circuit board vacuum lamination process according to claim 1, characterized in that: The steps for identifying the connected component contours of the bubble and wrinkled regions are as follows: An enhanced grayscale image is obtained by preprocessing the optical image on the substrate surface; Define a neighborhood window centered on each pixel in the enhanced grayscale image, and calculate the variance of grayscale values ​​of all pixels within the neighborhood window; If the gray value variance is greater than the texture variance threshold, the pixel is determined to be located in a potential defect area of ​​texture abrupt change and is marked as a foreground point; otherwise, it is marked as a background point, thereby generating a preliminary first binary image of the bubble and wrinkle abnormal areas. A morphological opening operation is performed on the first binary image to eliminate noise and smooth the region boundaries, resulting in a second binary image. Connected regions composed of all foreground points are then searched in the second binary image as candidate defect regions. For each candidate defect region, the corresponding local region is located by tracing back to the enhanced grayscale image, and the grayscale mean of the local region is calculated. If the grayscale mean is higher than the high grayscale threshold, it is determined to be a bubble region; if the grayscale mean is lower than the low grayscale threshold, it is determined to be a wrinkle region.

6. The method for multi-parameter detection and analysis of circuit board vacuum lamination process according to claim 1, characterized in that: The abnormal morphological parameters for calculating the contours of connected components include: Calculate the coordinates of the center point of the connected region contour as the location parameter of the defect; The area of ​​the connected component contour is obtained by counting the number of all pixels enclosed within the contour and multiplying this number by the actual area represented by a single pixel. Calculate the ratio of the length of the longer side to the shorter side of the smallest bounding rectangle of the connected region outline, and use it as its aspect ratio.

7. The method for multi-parameter detection and analysis of circuit board vacuum lamination process according to claim 4, characterized in that: The identification of the uniform degradation diffusion gradient and direction includes: For each grid cell in the dry film uniformity feature distribution matrix, calculate the difference of its dry film uniformity feature value with that of its adjacent grid cells along the row and column directions of the matrix to form the horizontal gradient matrix and the vertical gradient matrix. Traverse the horizontal and vertical gradient matrices and extract the grid sequence whose gradient values ​​have the same sign and are spatially continuous as the local change vector; Connect and merge spatially adjacent local change vectors with the same gradient sign to form a continuous path that reflects the monotonic change of eigenvalues, which serves as the change propagation path; If all signs of the local change vectors in the change propagation path are positive, it is determined to be positive diffusion; if all signs are negative, it is determined to be negative diffusion. The principal component direction of the set of grid cells covered by the change propagation path is defined as the main diffusion direction.

8. The method for multi-parameter detection and analysis of circuit board vacuum lamination process according to claim 7, characterized in that: The step of locating areas with deteriorated dry film adhesion uniformity includes: The grid cells whose dry film uniformity characteristic values ​​show a monotonically changing trend in each propagation path, and the grid cells that are directly adjacent to any grid cell in the path and whose dry film uniformity characteristic values ​​are within the range of characteristic values ​​at both ends of the path, are defined as the initial influence area. If any two regions in the initial affected area have adjacent or overlapping grid cells, these two regions will be merged to form a merged region, which will be identified as a region of deteriorated dry film adhesion uniformity.

9. The method for multi-parameter detection and analysis of circuit board vacuum lamination process according to claim 6, characterized in that: S5 specifically includes: Calculate the average nearest neighbor distance of all defect center points. If the average nearest neighbor distance is less than the expected nearest neighbor distance, it is determined to be a spatial clustering pattern; otherwise, it is determined to be a spatial dispersion pattern. Calculate the ratio of the mean absolute deviation of the defect area to its arithmetic mean, and use it as the area dispersion. Similarly, calculate the aspect ratio dispersion. If the area dispersion is less than the aspect ratio dispersion, it is judged as a high similarity pattern; otherwise, it is judged as a low similarity pattern. If the spatial distribution of defects is dispersed and the morphological parameters are of low similarity, then it is determined to be a random and occasional risk type. If the spatial distribution of defects is clustered and the morphological parameters are highly similar, then it is determined to be a systemic or localized risk type.

10. A multi-parameter detection and analysis method for circuit board vacuum lamination process according to claim 9, characterized in that: The process parameters that determine the quality risks include: Based on the type of abnormal defect risk and the spatial location of the deteriorated area of ​​dry film adhesion uniformity, the time interval corresponding to the deteriorated area during the lamination process is determined, and the time sequence data of process parameters within the time interval is extracted. Process parameters that continuously deviate from the normal baseline range in the extracted process parameter time series data are identified as abnormal process parameters. If the risk is systemic or localized, the parameter with the highest percentage of abnormal state duration within the corresponding time interval is selected as the main related parameter. If it is a random and occasional risk, then the parameters that exhibit intermittent pulses within the corresponding time interval in the abnormal process parameters are identified as the main related parameters; The key related parameters were identified as the process parameters that caused this quality risk.

Images