Image Processing-Based Method and System for Detecting Defects in Adhesive Application

Abstract

This application relates to the field of image processing technology, and discloses a method and system for detecting adhesive application defects based on image processing. The method includes: acquiring raw data by imaging the adhesive application area of ​​a PCB board using a narrow-band blue LED; selecting and identifying different adhesives using an adaptive color separation channel to obtain feature data; performing three-dimensional adhesive path modeling analysis based on the feature data to form a three-dimensional adhesive model; generating regional defect data by performing multi-level tolerance assessment on different functional areas based on the three-dimensional model; performing feature enhancement analysis on the defects using specific lighting techniques to form a feature map; and dynamically adjusting the output results using a temperature and humidity compensation algorithm. This application can adapt to different adhesive types, different PCB backgrounds, and different functional area requirements, and can consider environmental factors and batch differences in materials, improving the accuracy and reliability of detection and meeting the needs of high-quality production in the electronics manufacturing industry.

Description

Technical Field

[0001] This application relates to the field of image processing technology, and particularly to a glue application defect detection method and system based on image processing. Background Art

[0002] In the electronics manufacturing industry, glue application is a key process in connection, fixation, heat dissipation, etc. Its quality directly affects the performance and reliability of products. Traditional glue application quality inspection mainly relies on manual visual inspection, which has problems such as low efficiency, inconsistent standards, and inability to detect tiny defects. With the improvement of automation, detection methods based on machine vision are gradually applied to production lines. Images are collected by cameras and analyzed to identify defects in glue application. Existing detection technologies usually adopt fixed light sources and single recognition algorithms, use image segmentation and feature extraction technologies to detect the glue area, and compare with preset standards to determine whether it is qualified. These methods have improved the automation level and accuracy of detection to a certain extent.

[0003] However, there are still many deficiencies in the existing technology. First, for glues and PCB board backgrounds of different colors, a single image processing method is difficult to adapt to and is prone to misjudgment. Second, for defects such as tiny bubbles, uneven thin layer coverage, and tiny contamination, the detection rate of conventional imaging methods is low. Third, existing detection systems generally adopt a unified judgment standard and cannot conduct differential evaluation according to the specific requirements of different functional areas on the PCB board. Fourth, changes in temperature and humidity in the production environment will significantly affect the fluidity and curing characteristics of the glue, but existing detection methods rarely consider the influence of environmental factors on the detection results. Finally, the characteristic differences between different batches of glue will also lead to inconsistent detection results. These problems result in insufficient accuracy and reliability of the existing glue application defect detection technology in practical applications and cannot meet the quality requirements of high-precision electronic product manufacturing. Summary of the Invention

[0004] This application provides a glue application defect detection method and system based on image processing, which can adapt to different glue types, different PCB backgrounds, and different functional area requirements, and can consider environmental factors and material batch differences, improving the accuracy and reliability of detection and meeting the needs of high-quality production in the electronics manufacturing industry.

[0005] Firstly, this application provides an image processing-based method for detecting adhesive application defects. The method includes: illuminating the adhesive application area of ​​a PCB board in an electronics manufacturing production line using a narrow-band blue LED light source to acquire an image of the adhesive application on the PCB board, obtaining raw adhesive image data; based on the raw adhesive image data, identifying and processing different colored adhesives through adaptive color separation channel selection to obtain adhesive area feature data; and based on the adhesive area feature data, performing trajectory analysis on the adhesive distribution in a specific area of ​​the product through three-dimensional adhesive path modeling. A three-dimensional model of the adhesive, including its planar position and thickness distribution, is established. Based on this model, a multi-level tolerance assessment is used to differentiate the adhesive application quality in key functional areas, secondary functional areas, and non-functional areas of the PCB board, generating regional defect data. According to this regional defect data, feature enhancement analysis is performed on micro-bubbles, uneven thin-layer coverage, and adhesive contamination using specific angle side lighting and dual-source switching to form a defect feature map. Based on this defect feature map, a temperature and humidity compensation algorithm is used to dynamically adjust the parameters to account for differences in adhesive performance caused by changes in the production environment, outputting the defect judgment result.

[0006] Secondly, this application provides an image processing-based glue application defect detection system, the image processing-based glue application defect detection system comprising:

[0007] The acquisition module is used to illuminate the glue application area of ​​the PCB board in the electronic manufacturing production line using a narrow-band blue LED light source, acquire glue application images of the PCB board, and obtain raw glue image data.

[0008] The recognition module is used to identify different colors of glue based on the original glue image data by selecting the adaptive color separation channel, and obtain glue area feature data.

[0009] The analysis module is used to perform trajectory analysis on the glue distribution in a specific area of ​​the product based on the glue area feature data and through three-dimensional glue path modeling, and to establish a three-dimensional glue model that includes planar position and thickness distribution.

[0010] The judgment module is used to differentiate the glue application quality of key functional areas, secondary functional areas and non-functional areas of the PCB board based on the glue three-dimensional model and through multi-level tolerance evaluation, and generate regional defect data.

[0011] The enhancement module is used to perform feature enhancement analysis on microbubbles, uneven thin-layer coverage, and glue contamination based on the regional defect data, through specific angle side lighting and dual light source switching, to form a defect feature map;

[0012] The adjustment module is used to dynamically adjust the parameters of the adhesive performance differences caused by changes in the production environment based on the defect feature map and through a temperature and humidity compensation algorithm, and output the defect judgment result.

[0013] A third aspect of the present invention provides a computer device, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor invokes the instructions in the memory to cause the computer device to perform the above-described image processing-based glue application defect detection method.

[0014] A fourth aspect of the present invention provides a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the above-described image processing-based glue application defect detection method.

[0015] The technical solution provided in this application uses a narrow-band blue LED light source to precisely illuminate the adhesive application area of ​​the PCB board in the electronic manufacturing production line, achieving optimal contrast between different colored adhesives and the PCB board background, and significantly improving the quality of the original image data. Then, an adaptive color separation channel selection technology is employed to automatically select the optimal recognition channel based on the spectral characteristics of different colored adhesives, overcoming the limitation of traditional single-channel methods that cannot adapt to multiple adhesive types. The three-dimensional adhesive path modeling process fully considers the three-dimensional morphological characteristics of the adhesive, analyzing not only planar distribution but also thickness information, extending detection from two-dimensional to three-dimensional, and better meeting practical application needs. A multi-level tolerance evaluation mechanism sets differentiated detection standards for different functional areas on the PCB board, avoiding excessive rejection or missed detection caused by a one-size-fits-all approach. Specific angle side lighting and dual-source switching technology significantly improve the detection capability of difficult-to-detect defects such as micro-bubbles, uneven thin-layer coverage, and micro-adhesive droplet contamination, achieving a detection accuracy on the order of 0.1mm. The temperature and humidity compensation algorithm innovatively incorporates environmental factors, automatically adjusting the judgment parameters and solving the detection fluctuation problem caused by environmental changes. This invention organically integrates artificial intelligence algorithms into the glue defect detection process. Particularly in the circular feature enhancement filtering, multi-level tolerance assessment, and temperature and humidity compensation stages, the algorithmic features significantly improve the system's adaptability and robustness, enabling the detection process to adaptively handle complex scenarios under different working conditions. The comprehensive application of these technical features not only greatly improves the accuracy and consistency of glue application defect detection, but more importantly, it can adapt to various application scenarios in the electronics manufacturing industry, including key application areas such as power device heat dissipation, chip sealing, and waterproofing and dustproofing, greatly improving product quality and reliability. Furthermore, by establishing a batch feature database, this invention solves the problem of fluctuations in detection standards caused by different batches of glue, providing a stable and consistent quality control capability for the production process and improving detection efficiency. Attached Figure Description

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

[0017] Figure 1 This is a schematic diagram of an embodiment of the glue application defect detection method based on image processing in this application.

[0018] Figure 2 This is a schematic diagram of an embodiment of the glue application defect detection system based on image processing in this application.

[0019] Figure 3 This is a schematic block diagram of the structure of the computer device in an embodiment of the present invention. Detailed Implementation

[0020] This application provides a method and system for detecting glue application defects based on image processing. The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data used can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0021] For ease of understanding, the specific process of the embodiments of this application is described below. Please refer to [link / reference]. Figure 1 One embodiment of the glue application defect detection method based on image processing in this application includes:

[0022] Step S101: Illuminate the PCB board adhesive application area in the electronic manufacturing production line with a narrow-band blue LED light source, collect the adhesive application image of the PCB board, and obtain the original adhesive image data.

[0023] Step S102: Based on the original glue image data, different colored glues are identified and processed through adaptive color separation channel selection to obtain glue area feature data.

[0024] Step S103: Based on the adhesive area feature data, perform trajectory analysis on the adhesive distribution in a specific area of ​​the product through three-dimensional adhesive path modeling, and establish a three-dimensional adhesive model that includes planar position and thickness distribution.

[0025] Step S104: Based on the three-dimensional model of the adhesive, the adhesive application quality of the key functional area, secondary functional area and non-functional area of ​​the PCB board is differentiated by multi-level tolerance assessment, and regional defect data is generated.

[0026] Step S105: Based on the defect data of different regions, perform feature enhancement analysis on microbubbles, uneven thin-layer coverage and glue contamination by using side lighting at a specific angle and switching between dual light sources to form a defect feature map.

[0027] Step S106: Based on the defect feature map, dynamically adjust the parameters of the adhesive performance differences caused by changes in the production environment through the temperature and humidity compensation algorithm, and output the defect judgment result.

[0028] It is understood that the executing entity of this application can be an image processing-based glue application defect detection system, or it can be a terminal or a server; no specific limitation is made here. This application's embodiments use a server as an example for illustration.

[0029] Specifically, the process begins with precise lighting and image acquisition. A narrow-band blue LED light source illuminates the adhesive-coated area of ​​the PCB board. This narrow-band blue LED light source is set within the 470nm-480nm wavelength range, providing optimal contrast with commonly used thermal grease and epoxy adhesive. The light source is positioned at a 30° incident angle to avoid PCB surface reflections interfering with image quality. A high-resolution CMOS camera (12 megapixels) is fixed 300mm behind the adhesive coating station on the SMT production line, ensuring coverage of the entire PCB board area without interfering with production. During acquisition, the camera shutter speed is set to 1 / 1000 second, ensuring clear images even during production line movement. Acquisition triggering is synchronized with the production line conveyor belt, precisely triggering the capture when the PCB board arrives, forming raw adhesive image data. Adhesive area recognition processing is then performed, using an adaptive color separation channel to select and identify different colored adhesives. A standard grayscale reference block is preset at the edge of the PCB board, and the lighting compensation coefficient is calculated by comparing the actual brightness of the reference block with the standard value. The original image is multiplied by the compensation coefficient to generate a standardized illumination image, which is then decomposed into three RGB channels. The signal-to-noise ratio (SNR) of each channel is calculated. For example, in an electronics production line, white epoxy adhesive on a green PCB background has a SNR of 12.6 for the G channel, which is much higher than the 4.8 for the R channel and the 7.2 for the B channel. Therefore, the G channel is selected as the adhesive identification channel. The PCB silkscreen and copper foil background interference areas in the selected channel are masked to generate a background-removed image. For the transparent silicone area, reflections are eliminated by adjusting the polarization filter parameters to generate a boundary-enhanced image. The position coordinates, area, and boundary data of the adhesive area are extracted from this image to form the adhesive area feature data.

[0030] Subsequently, 3D adhesive path modeling was conducted, and trajectory analysis of adhesive distribution in specific areas of the product was performed based on adhesive area feature data. PCB board CAD drawings were imported from the production design database to extract theoretical adhesive path trajectory information and generate a basic adhesive path outline. Spatially registration was performed between the positional information from the adhesive area feature data and the basic adhesive path outline to form an actual adhesive path distribution map. The actual adhesive path distribution map was then categorized and labeled, dividing the PCB board into mandatory coating areas (e.g., CPU heat dissipation interfaces), prohibited coating areas (e.g., around connector contacts), and optional areas (non-critical areas). Characteristic parameter tables were established for different adhesive types, recording the shrinkage rate parameters (6%-8%) of epoxy resin adhesives and the leveling parameters of silicone adhesives, thus constructing an adhesive physicochemical property library. Based on the specific distribution requirement of "thicker in the middle and thinner at the edges" for thermal grease in server heatsink installation, thickness distribution curves were added to the actual adhesive path distribution map, ultimately forming a 3D adhesive model containing planar position and thickness information.

[0031] Based on the resulting 3D model of the adhesive, a multi-level tolerance assessment was conducted to differentiate the adhesive application quality across different functional areas of the PCB board. The PCB board functional areas were divided into three levels: a critical functional area (power device heat dissipation surface and waterproof sealing line), a secondary functional area (fixed bracket and auxiliary heat dissipation area), and a non-functional area (marking area). Three levels of tolerance standards were established for each: critical functional area required adhesive coverage of no less than 95% and thickness error of no more than ±10%; secondary functional area required coverage of no less than 85% and thickness error of no more than ±20%; non-functional area primarily assessed for adhesive overflow and contamination, with an overflow area not exceeding 0.5 square meters. Based on the distribution data in the 3D model of the adhesive, the actual adhesive coverage and thickness deviation of each area are calculated. For example, the coverage deviation is obtained by measuring the ratio of the actual adhesive coverage area to the theoretical area in key functional areas, and the thickness deviation is obtained by calculating the difference between the thickness and the standard thickness. Finally, the data is compared with the corresponding standard to generate defect data for each area.

[0032] To accurately identify various minute defects, feature enhancement analysis was performed using side lighting at specific angles and dual-source switching technology. For microbubbles, side lighting at angles of 30°-60° was used to enhance the reflectivity of the bubble edges, generating enhanced bubble images. A circular feature enhancement filter was then applied to highlight the outline of bubbles with a diameter less than 0.3 mm. For areas with uneven thin-layer coverage, dual-source illumination was used, sequentially capturing images with a 0° vertical light source and a 45° side light source. The difference between the two sets of images was calculated to extract areas of varying reflectivity. High-resolution local imaging was performed on glue-contaminated areas to identify tiny glue droplets as small as 0.1 mm × 0.1 mm. These feature data were then integrated to form a defect feature map.

[0033] A temperature and humidity compensation algorithm dynamically adjusts adhesive performance differences caused by variations in the production environment. Real-time monitoring of factory temperature and humidity parameters is performed, and temperature-viscosity and humidity-curing rate curves for different adhesive types are retrieved to obtain theoretical performance parameters for the adhesive under the current environment. These parameters are compared with performance parameters under a standard environment (23℃ / 50%RH) to calculate environmental factor compensation coefficients. Leveling deviations of silicone adhesives are mitigated at high temperatures, and continuity defects of epoxy adhesives are mitigated at low temperatures, generating environmentally compensated defect evaluation values. Simultaneously, batch-to-batch variation factors are considered, and batch characteristic databases are used to record the characteristics of each batch of adhesive to generate batch compensation coefficients, ultimately outputting accurate defect judgment results.

[0034] In this embodiment, a narrow-band blue LED light source is used to precisely illuminate the adhesive application area of ​​the PCB board in the electronic manufacturing production line, achieving optimal contrast between different colored adhesives and the PCB board background, significantly improving the quality of the original image data. Then, an adaptive color separation channel selection technology is employed to automatically select the optimal recognition channel based on the spectral characteristics of different colored adhesives, overcoming the limitation of traditional single-channel methods that cannot adapt to multiple adhesive types. The three-dimensional adhesive path modeling process fully considers the three-dimensional morphological characteristics of the adhesive, analyzing not only planar distribution but also thickness information, extending detection from two-dimensional to three-dimensional, better meeting practical application needs. A multi-level tolerance evaluation mechanism sets differentiated detection standards for different functional areas on the PCB board, avoiding excessive rejection or missed detection caused by a one-size-fits-all approach. Specific angle side lighting and dual-source switching technology significantly improve the detection capability for difficult-to-detect defects such as micro-bubbles, uneven thin-layer coverage, and micro-adhesive droplet contamination, achieving a detection accuracy on the order of 0.1 mm. The temperature and humidity compensation algorithm innovatively incorporates environmental factors, automatically adjusting judgment parameters and solving the detection fluctuation problem caused by environmental changes. This invention organically integrates artificial intelligence algorithms into the glue defect detection process. Particularly in the circular feature enhancement filtering, multi-level tolerance assessment, and temperature and humidity compensation stages, the algorithmic features significantly improve the system's adaptability and robustness, enabling the detection process to adaptively handle complex scenarios under different working conditions. The comprehensive application of these technical features not only greatly improves the accuracy and consistency of glue application defect detection, but more importantly, it can adapt to various application scenarios in the electronics manufacturing industry, including key application areas such as power device heat dissipation, chip sealing, and waterproofing and dustproofing, greatly improving product quality and reliability. Furthermore, by establishing a batch feature database, this invention solves the problem of fluctuations in detection standards caused by different batches of glue, providing a stable and consistent quality control capability for the production process and improving detection efficiency.

[0035] In one specific embodiment, the process of performing step S101 may specifically include the following steps:

[0036] (1) Set the narrowband blue LED light source to a wavelength range of 470nm-480nm, and place the narrowband blue LED light source above the PCB board surface at an incident angle of 30° to avoid PCB surface reflection interference;

[0037] (2) High-definition imaging of the PCB board is performed at a fixed position 300mm behind the glue coating station of the SMT production line, covering a maximum range of 350mm×280mm of the PCB board.

[0038] (3) Set the camera shutter speed to 1 / 1000 second to ensure that clear images are obtained while the production line is in motion;

[0039] (4) The PCB board is detected by photoelectric sensor to accurately trigger the image acquisition system and take full-coverage pictures of the glue application area of ​​the PCB board;

[0040] (5) Automatically switch preset lighting parameters according to different product models to enhance the directional lighting of the glue application area on the PCB board;

[0041] (6) Encode and store the high-definition image data of the PCB board glue application area to generate the original glue image data.

[0042] Specifically, the narrowband blue LED light source was set to a wavelength range of 470nm-480nm using the image acquisition system. This specific wavelength range was chosen based on spectral test results, which showed that this wavelength range produces the best contrast with materials commonly used in electronic manufacturing, such as epoxy resin and silicone. The absorption coefficient of the adhesive at this wavelength differs significantly from that of the PCB substrate material, forming a clear spectral characteristic boundary. The narrowband blue LED light source was positioned above the PCB surface at a 30° incident angle. This angle avoids specular reflections from the PCB surface directly entering the camera lens, reducing saturation issues in highlight areas. When light is incident at a 30° angle, most of the reflected light deviates from the camera's optical axis, while the scattering characteristics of the adhesive material scatter the incident light uniformly in all directions, thus forming good material identification features in the image.

[0043] The image acquisition equipment is fixed 300mm behind the adhesive coating station on the SMT production line. This distance was chosen after considering factors such as field of view coverage, space limitations, and production interference. A high-definition industrial camera with an appropriate focal length lens is used to ensure sufficient spatial resolution by covering a maximum inspection area of ​​350mm × 280mm on the PCB board at a working distance of 300mm. The pixel density reaches over 20 pixels per millimeter, capable of resolving minute adhesive features as small as 0.05mm, meeting the requirements for detecting tiny air bubbles and minor contaminants. The camera uses a global shutter CMOS sensor, avoiding the distortion problems encountered when imaging moving objects with a rolling shutter.

[0044] The shutter speed was set to 1 / 1000 second, a parameter optimized for the conveyor belt speed of the SMT production line. When the production line operates at a speed of 25 pieces / minute (approximately 0.4 pieces / second), the displacement of the PCB board during exposure does not exceed 0.4mm, far below the minimum detection feature size, thus ensuring image sharpness. To balance the exposure, the light source power was correspondingly increased, reaching an illuminance of 100,000 lux, compensating for underexposure caused by the shortened shutter speed. The image acquisition trigger mechanism uses a photoelectric sensor to detect the PCB board's arrival signal. The sensor is installed at an appropriate position on the conveyor belt. When the leading edge of the PCB board blocks the photoelectric sensor's optical path, a TTL level transition trigger signal is generated. This signal, after being processed by a delay compensation circuit, precisely controls the camera's exposure time, compensating for the spatial difference between the sensor position and the shooting position, ensuring that the PCB board is fully within the center of the field of view before triggering the shot, avoiding partial truncation or deviation from the field of view. The trigger accuracy is controlled within ±1mm, ensuring a stable and consistent detection area.

[0045] The automatic switching of preset lighting parameters based on different product models is achieved through communication with the production line's MES system. It retrieves the current PCB model information from the production scheduling database, queries the lighting parameter database, and automatically calls upon the corresponding combination of light source brightness, color temperature, and illumination angle. For black PCBs, the light source brightness is increased by 30%; for white or silver-plated PCBs, the light source brightness is reduced by 20% and a polarizing filter is added to reduce highlights; for transparent adhesives, a symmetrical dual-sided light source configuration is used to enhance edge contrast. The lighting scheme switching process is fully automated, with an average switching time of no more than 2 seconds, without affecting production cycle time.

[0046] During the encoding and storage of high-definition image data of the PCB board adhesive application area, the original images are first preprocessed, including dark field correction and flat field correction, to eliminate inherent noise from the camera sensor and lens vignetting effects. The image data is saved using a lossless compression format, such as PNG, to ensure no loss of image quality. Each image is tagged with metadata, including the PCB board serial number, product model, shooting timestamp, and environmental parameters, facilitating subsequent analysis and batch traceability. The original adhesive image data has a resolution of 4096×3072 pixels, a bit depth of 12 bits, and a data size of approximately 24MB per image, and is stored and managed using a computer vision processing platform.

[0047] Taking the inspection of a certain type of power amplifier PCB board as an example, after the adhesive is applied, the PCB board is conveyed to the inspection position by a conveyor belt. The photoelectric sensor detects the leading edge of the PCB board and generates a trigger signal. The system identifies the current production model as "PA-2000" and automatically switches to the preset illumination scheme 3, which includes a combination of 75% light source brightness and a polarizing filter. The camera captures the image at a shutter speed of 1 / 1000 second. After dark field and flat field correction, the raw image is converted into a standardized 12-bit PNG format for storage and appended with metadata tags of "PA-2000-S12345-20250313-093045-T24C-RH55%", completing the acquisition process of the raw adhesive image data.

[0048] In one specific embodiment, the process of performing step S102 may specifically include the following steps:

[0049] (1) Set a standard grayscale reference block at the edge of the PCB board, and calculate the current illumination compensation coefficient by comparing the difference ratio between the actual brightness value of the reference block and the standard value;

[0050] (2) Multiply the original glue image data by the illumination compensation coefficient to generate a standardized illumination image, and decompose the standardized illumination image into single-channel images of the three RGB channels;

[0051] (3) By calculating the signal-to-noise ratio index of a single-channel image, the single channel with the highest contrast is determined as the glue recognition channel;

[0052] (4) Mask the PCB silkscreen and copper foil background interference areas in the glue recognition channel to generate a background-removed image;

[0053] (5) For transparent or semi-transparent silicone areas in the background removal image, the reflection is eliminated by adjusting the polarization filter parameters to generate a boundary enhancement image;

[0054] (6) Apply shape regularity and edge transition characteristics analysis to the boundary enhancement image, extract the position coordinates, area values ​​and boundary data of the glue application area, and form glue area feature data.

[0055] Specifically, standard grayscale reference blocks are pre-set on the edge of the PCB board. These reference blocks consist of four small squares with different grayscale levels (25%, 50%, 75%, and 100%), each with an area of ​​5mm × 5mm. They are made of a special wear-resistant material and fixed to the edge of the PCB conveyor tray without interfering with the PCB board itself. Each grayscale reference block has a pre-calibrated standard brightness value under standard lighting conditions, which is stored in the detection system database. The illumination compensation coefficient is calculated using the following formula:

[0056]

[0057] in, It is the illumination compensation coefficient. This refers to the number of reference blocks. It is the first Standard brightness values ​​for each reference block It is the first The actual measured brightness values ​​of each reference block under the current lighting conditions are used to compensate for the impact of changes in ambient lighting on the image.

[0058] The original glue image data is standardized by multiplying it by an illumination compensation coefficient to generate a standardized illumination image, which is then decomposed into single-channel images with three RGB channels. To determine the optimal glue recognition channel, the signal-to-noise ratio (SNR) of each single-channel image is calculated using the following formula:

[0059]

[0060] in, It is a passage The signal-to-noise ratio (SNR) value (which can be R, G, or B). This is the average brightness of the glued area in that channel. This is the average brightness of the PCB background area in this channel. and These represent the standard deviations of the glue area and the PCB background area in this channel, respectively. System Selection The channel with the highest value is used as the glue recognition channel.

[0061] The PCB silkscreen and copper foil background interference areas in the adhesive recognition channel are masked. A background area mask is created using pre-stored PCB design data to generate a background-removed image. For transparent or semi-transparent silicone areas in the background-removed image, polarized light filter parameters are adjusted to eliminate reflective interference. Shape regularity and edge transition characteristics are analyzed on the boundary enhancement image to extract features of the adhesive application area.

[0062] Taking the application of thermal paste in power amplifiers in the electronics manufacturing industry as an example, a PCB board contains multiple power devices that require thermal grease application. After acquiring the original image, the system detected that the actual brightness values ​​of four grayscale reference blocks were 61, 124, 186, and 243, while the standard values ​​should be 64, 128, 192, and 256, resulting in a lighting compensation coefficient of 1.053. Multiplying each pixel value of the original image by 1.053 yields a standardized image, which is then decomposed into RGB three channels. The signal-to-noise ratio (SNR) of the three channels is calculated, showing that the R channel is 5.24, the G channel is 8.76, and the B channel is 3.12. Therefore, the G channel is selected as the paste recognition channel. Through mask processing and polarization filter parameter adjustment, the position coordinates, area values, and boundary data of eight thermal paste areas are successfully extracted, forming paste area feature data.

[0063] In one specific embodiment, the process of executing step S103 may specifically include the following steps:

[0064] (1) Import CAD drawings of PCB products from the production design database, extract theoretical glue path information, and generate basic glue path outline diagram;

[0065] (2) Spatial registration of the location coordinates, area values ​​and boundary data in the glue area feature data with the basic glue path outline map to form the actual glue path distribution map;

[0066] (3) The actual glue distribution map is classified and marked into three types of glue application areas: mandatory application area, prohibited application area and optional application area;

[0067] (4) Establish characteristic parameter tables for different types of adhesives, record the shrinkage parameters of epoxy resin adhesives and the leveling parameters of silicone adhesives, and construct a library of physicochemical properties of adhesives;

[0068] (5) Based on the thermal grease distribution conditions in the installation of the server heat sink, add a thickness distribution curve to the actual adhesive path distribution diagram, wherein the thermal grease distribution condition is thick in the middle and thin at the edges;

[0069] (6) Combine the actual glue path distribution diagram with the thickness distribution curve to form a three-dimensional glue model containing planar position and thickness information.

[0070] Specifically, CAD drawings of PCB products are imported from the production design database. The CAD interface module reads Gerber files or ODB++ format PCB design data generated by electronic product design software, extracting layer information related to adhesive application. During import, the vector data in the PCB design file is parsed into point sets and line segment sets to identify the geometry of the adhesive application area, including linear adhesive paths, dotted adhesive dots, and planar adhesive layer areas. Through vector-to-raster conversion, these geometries are converted into pixelated trajectory maps while preserving coordinate and dimensional information, ultimately generating a basic adhesive path outline map. This map contains the ideal adhesive distribution path and area boundaries specified in the design phase.

[0071] Spatial registration of the glue area feature data with the basic glue path outline is a crucial step in handling positional deviations during actual production. Since the position and angle of the PCB board on the conveyor belt may have slight deviations, a rigid transformation algorithm is required for spatial registration. Specifically, key feature points, including glue path corners and endpoints, are first extracted from the glue area feature data, and then matched with corresponding points in the basic glue path outline. The optimal rotation matrix and translation vector are calculated using the least squares method to transform the coordinate system of the actual glue area to the design reference coordinate system. This transformation includes translation and rotation operations, handling positional deviations of ±2mm and rotation errors of ±3°. After the transformation, the position coordinates, area values, and boundary data from the glue area feature data are mapped to the design reference coordinate system, forming an actual glue path distribution map in the same coordinate system as the basic glue path outline.

[0072] Classifying and labeling the actual adhesive path distribution map involves dividing different adhesive application areas according to functional requirements. Mandatory areas are those requiring high adhesive coverage, such as CPU heat dissipation interfaces and power device heat dissipation surfaces—critical heat conduction areas. No-adhesion areas are those where adhesive is absolutely prohibited, such as connector contacts and gold finger interfaces—electrical contact areas. Optional areas are non-critical areas where adhesive application requirements are not strict. The labeling process uses a region overlap analysis algorithm to compare the actual adhesive path distribution map with a predefined functional area layer. Based on the overlap, each pixel is assigned a corresponding region type label, generating a classification and labeling map containing these three types of region tags.

[0073] Establishing characteristic parameter tables for different adhesive types is crucial for accurately predicting the morphological changes of the adhesive after application. Epoxy resin adhesives have a certain shrinkage rate, typically shrinking by 6%-8% in volume during curing, which leads to a reduction in adhesive thickness and coverage area. Silicone-based materials, on the other hand, exhibit leveling properties, spreading outwards under their own gravity and surface tension after application. The characteristic parameter tables record key parameters such as viscosity, flowability, and curing time of the adhesive under different temperature conditions. These parameters are derived from technical specifications provided by material suppliers and laboratory test data. By establishing a mathematical model, the final distribution pattern of the adhesive can be predicted based on the initial application state, allowing for optimization and adjustment of design parameters. Due to the specific requirements of thermal grease in server heatsink installations, a thickness distribution curve needs to be added to the actual adhesive path distribution diagram. The heat dissipation interface of high-power components such as server CPUs requires the thermal grease to have a "thicker in the middle and thinner at the edges" distribution. This is because when the heatsink is compressed, the grease flows outwards from the center, forming optimal heat conduction contact. The thickness distribution curve is generated using a Gaussian function model, with the thickness being greatest at the center and gradually decreasing towards the edges. Matrix operations are used to map the thickness parameter to each pixel of the actual adhesive path distribution map, generating a three-dimensional height matrix. The values ​​in the height matrix represent the adhesive thickness at each location, typically set to 0.2mm-0.3mm in the central area and gradually thinning to 0.05mm-0.1mm at the edges.

[0074] The final step in generating a 3D model of the adhesive is to combine the actual adhesive distribution map with the thickness distribution curve. This process integrates two-dimensional planar position information and thickness information in the height direction into a 3D data structure. For each pixel marked as an adhesive area in the actual adhesive distribution map, a corresponding thickness value is assigned, forming 3D point cloud data. This point cloud data is converted into a surface mesh model through meshing, recording the complete 3D morphology of the adhesive distribution. The 3D model of the adhesive not only contains the planar shape and area information of the adhesive but also the thickness distribution information, enabling it to more accurately reflect the three-dimensional form of the adhesive in practical applications.

[0075] Taking the application of thermal paste for a certain type of server motherboard CPU as an example, the CAD file imported from the production design database shows that the CPU heat dissipation interface is a 40mm × 40mm square area, and the theoretical adhesive path is a center dot application. Adhesive area characteristic data shows that the center position of the actual applied adhesive dots is offset by 1.2mm from the designed position and rotated 1.5° clockwise. After correction using a spatial registration algorithm, the actual adhesive path distribution map shows that the adhesive dots cover an area of ​​38mm × 38mm. Area classification labeling divides the CPU heat dissipation interface into a core mandatory application area (central 30mm × 30mm) and an edge transition area (a 5mm wide annular area on the periphery). According to the characteristic parameter table, the viscosity of the thermal grease used at 25℃ is 180,000 cP, the leveling coefficient is 0.85, and the shrinkage rate is 2%. The thickness distribution curve is designed with a maximum thickness of 0.25mm at the center and a minimum thickness of 0.08mm at the edge, forming a smooth raised shape. The generated 3D model of the adhesive describes the planar distribution and thickness variation of the CPU thermal paste.

[0076] In one specific embodiment, the process of executing step S104 may specifically include the following steps:

[0077] (1) Divide the PCB board into functional areas, and divide the PCB board into key functional areas such as the heat dissipation surface of power devices and waterproof sealing line, secondary functional areas such as the fixed bracket and auxiliary heat dissipation area, and non-functional areas such as the marking area.

[0078] (2) Establish a first-level tolerance standard for key functional areas and set a judgment threshold of glue coverage of not less than 95% and thickness error of not more than ±10%.

[0079] (3) For secondary functional areas, a second-level tolerance standard is formulated, and a judgment threshold is set that the glue coverage rate is not less than 85% and the thickness error is not more than ±20%.

[0080] (4) A third-level tolerance standard is formulated for non-functional areas, mainly to detect glue overflow pollution, and a judgment threshold is set for the overflow area not exceeding 0.5 square millimeters;

[0081] (5) Based on the glue distribution data in the glue three-dimensional model, compare the first-level tolerance standard, the second-level tolerance standard and the third-level tolerance standard, and calculate the glue application deviation value of each functional area;

[0082] (6) Compare the glue application deviation value of each functional area with the tolerance standard of the corresponding functional area to generate regional defect data including glue missing, glue overflow and glue thickness abnormality.

[0083] Specifically, the PCB board is functionally divided. This division is based on the actual working requirements and reliability requirements of electronic products, dividing the PCB board into three categories: critical functional areas, secondary functional areas, and non-functional areas. Critical functional areas include heat dissipation surfaces for power devices and waterproof sealing lines; these areas directly affect the core performance and reliability of the product. Secondary functional areas include mounting brackets and auxiliary heat dissipation areas; these areas have some impact on the product but are not fatal. Non-functional areas mainly consist of marking areas and other parts that do not directly affect the product's functionality. The area division process involves reading the functional layer information from the PCB design file, combining it with the product functional requirements document, creating a functional area mask matrix, and representing different areas with different label values ​​on an image processing platform. A differentiated tolerance standard system is established for each functional area. The critical functional area is subject to the first-level tolerance standard, which has the strictest requirements, setting a threshold of adhesive coverage of no less than 95% and thickness error of no more than ±10%. Coverage refers to the ratio of the actual adhesive coverage area to the design requirement coverage area, expressed as a percentage; thickness error refers to the deviation ratio between the actual adhesive thickness and the design requirement thickness. These stringent standards ensure efficient heat dissipation and waterproofing performance for critical components such as the CPU and power amplifier. Secondary functional areas are subject to Level 2 tolerance standards, which are relatively lenient, setting thresholds for adhesive coverage of at least 85% and thickness error not exceeding ±20%. Non-functional areas are subject to Level 3 tolerance standards, primarily focusing on adhesive overflow contamination, setting a threshold for overflow area not exceeding 0.5 square millimeters to prevent adhesive contamination of electrical contact areas such as gold fingers.

[0084] Based on the adhesive distribution data in the 3D model of the adhesive, it is necessary to calculate the adhesive application deviation value for each functional area. For key functional areas, the coverage deviation is calculated first:

[0085]

[0086] in, It is the deviation value of glue coverage in key functional areas. This is the actual area covered by the glue. This is the required glue coverage area as specified in the design. Thickness deviation is also calculated.

[0087]

[0088] in, It is the glue thickness deviation value in the key functional areas. This is the actual measured thickness of the glue. It is the glue thickness specified in the design.

[0089] For secondary functional areas, a similar formula is used to calculate the coverage deviation. and thickness deviation For non-functional areas, calculate the spillover pollution area:

[0090]

[0091] in, It is the area of ​​glue overflow in the non-functional area. Indicates non-functional areas. It is an indicator function, when the point The value is 1 if there is glue present, otherwise it is 0.

[0092] The calculated glue application deviation values ​​for each functional area are compared with the corresponding tolerance standards to determine defects. For critical functional areas, if... If the coverage is less than 95%, it is considered a defect due to insufficient glue coverage; if | If the thickness exceeds 10%, it is considered a thickness abnormality defect. For secondary functional areas, if... If the coverage is less than 85%, it is considered a defect due to insufficient glue coverage; if | If the thickness exceeds 20%, it is considered a thickness abnormality defect. For non-functional areas, if... > If the result is not found, it is determined to be a glue overflow contamination defect. All defect determination results are integrated into regional defect data, including defect type, location, and severity information.

[0093] Taking the camera module mounting area of ​​a smartphone motherboard as an example, this area includes the image processing chip heat dissipation surface (critical functional area), the module mounting bracket (secondary functional area), and the surrounding nameplate markings (non-functional area). 3D model analysis of the adhesive shows that the actual adhesive coverage area of ​​the image processing chip heat dissipation surface is 96%. The design requires a coverage area of ​​102 The calculated coverage deviation was 94.1%, lower than the first-level tolerance standard of 95%, and was therefore judged as an insufficient adhesive coverage defect. The actual adhesive thickness in the center area of ​​the chip heat dissipation surface was 0.22mm, while the design requirement was 0.20mm. The calculated thickness deviation was +10%, which just reached the ±10% tolerance limit and was not judged as a defect. The actual adhesive coverage area in the module fixing bracket area was 85... The design requires an area of ​​98 square meters. The coverage deviation was 86.7%, which meets the second-level tolerance standard. A small amount of glue overflow was observed in the nameplate marking area, with a measured overflow area of ​​0.3 square meters. It did not exceed 0.5 The third-level tolerance standard does not classify it as a defect. Comprehensive analysis generates regional defect data, including insufficient glue coverage on the heat dissipation surface of the image processing chip. The defect is located in the chip edge area, and the severity of the defect is minor (coverage is close to the judgment threshold).

[0094] In one specific embodiment, the process of executing step S105 may specifically include the following steps:

[0095] (1) For the micro bubble area in the sub-regional defect data, set a specific angle side lighting with a light source angle of 30°-60° to enhance the reflective properties of the bubble edge and generate bubble enhanced image;

[0096] (2) Apply a circular feature enhancement filter to the bubble enhancement image to highlight the outline of tiny bubbles with a diameter of less than 0.3 mm and obtain bubble feature data;

[0097] (3) For the uneven thin-layer coverage area in the sub-regional defect data, dual light source switching illumination is performed. The 0° vertical light source and the 45° side light source are used for illumination in sequence to obtain two sets of illumination images.

[0098] (4) Calculate the difference map of the two sets of lighting images, extract the areas with obvious differences in reflectivity, and form thickness unevenness feature data;

[0099] (5) For glue contamination areas in the regional defect data, high-resolution local magnification imaging and texture analysis are used to identify tiny glue droplets with a size as small as 0.1mm×0.1mm and generate contamination area feature data;

[0100] (6) Integrate bubble feature data, thickness unevenness feature data and contaminated area feature data to form a defect feature map containing defect type, location information and severity.

[0101] Specifically, for the microbubble areas in the regional defect data, a specific angle of side lighting is set, with the light source angle selected within the range of 30°-60°. This angle range is the optimal incident angle verified by optical experiments, enabling a significant reflective effect at the bubble edges. Bubbles form an interface between air and glue in the glue; when light is incident at a specific angle, the bubble edges produce significant light reflection, forming a bright ring. In practice, an auxiliary light source located at an angle of 30°-60° is added to the existing front light source. The brightness of the main light source remains constant, and the intensity of the auxiliary light source is adjusted to 80% of that of the main light source, resulting in an image with enhanced bubble edges. This special lighting method utilizes the light reflection characteristics of the bubble surface, making microbubbles, which are difficult to see under normal lighting, clearly visible in the image, thus generating a bubble-enhanced image.

[0102] Applying a circular feature enhancement filter to an enhanced bubble image further highlights the characteristics of tiny bubbles. The circular feature enhancement filter is a convolutional kernel specifically designed to enhance circular structures in an image, its design principle tailored to the geometric properties of bubbles. The core of the filter is a circular convolutional template, mathematically expressed as a two-dimensional function of center suppression and circular enhancement. During filtering, this filter is convolved with the image, enhancing the circular edge structures that match bubble characteristics while suppressing other non-circular structures. The filter parameters are optimized to specifically enhance tiny bubbles with diameters less than 0.3 mm, which are easily overlooked in traditional detection methods but significantly impact product performance. After filtering, the enhanced image is binarized and connected component analysis is performed to extract feature parameters such as bubble position, diameter, and area, forming bubble feature data.

[0103] For areas with uneven thin-layer coverage in the regional defect data, a dual-light source switching illumination technique was used for analysis. This technique utilizes the differences in reflectivity of adhesives of different thicknesses to light at different incident angles. First, a 0° vertical light source is used for illumination, primarily reflecting the smoothness and brightness uniformity of the adhesive surface. Then, a 45° side light source is used, which better displays the surface tilt caused by thickness variations. Both light sources use the same wavelength and brightness parameters to ensure comparable image brightness levels. By rapidly switching the light sources and taking separate images, two sets of illumination images of the same area under different lighting conditions are obtained. This process is precisely synchronized by a controller to ensure that the positions of the two images are completely consistent. Calculating the difference between the two sets of illumination images is the core step in identifying uneven thin-layer coverage. Pixel-level differences are calculated between the vertical and side-light images to obtain a difference map reflecting the differences in surface characteristics. In areas with uniform thickness, the differences in images produced by different light source angles are small; however, in areas with uneven thickness, due to changes in the surface tilt angle, the difference in reflectivity under the two light sources increases significantly. The difference map is segmented using a threshold process to extract regions where the reflectance difference exceeds a preset threshold. These regions typically correspond to locations where the glue thickness varies significantly. Morphological processing and connected component analysis are then performed on the extracted regions to record their location, area, shape, and other characteristic parameters, forming thickness unevenness feature data.

[0104] For glue contamination areas in the regional defect data, high-resolution local magnification imaging and texture analysis techniques are employed. High-resolution local magnification imaging involves secondary imaging of the identified potential contamination areas using a higher-magnification optical system to improve spatial resolution to the micrometer level, enabling the capture of tiny glue droplets that are difficult to detect with the naked eye. Texture analysis extracts texture features from the magnified images, including gray-level co-occurrence matrix features and local binary pattern features. These features effectively distinguish the texture differences between the glue and the PCB board surface. Through feature classification algorithms, tiny glue droplets as small as 0.1mm × 0.1mm can be accurately identified. The contours of the identified droplets are extracted, and their location coordinates, area, and shape parameters are recorded to generate contamination area feature data.

[0105] Integrating bubble feature data, thickness unevenness feature data, and contaminated area feature data is the process of forming a comprehensive defect feature map. The integration process first involves spatial registration of the various feature data types to ensure they are represented in the same coordinate system. Then, a multi-layered defect feature data structure is created, with each layer corresponding to a defect type and containing the defect's location information, geometric parameters, and severity. Defect severity is comprehensively evaluated based on the defect's area, quantity, and location importance. The resulting defect feature map is a structured dataset that details the type, location, and severity of all adhesive application defects on the PCB board, providing comprehensive information for subsequent defect analysis and processing.

[0106] Taking the production of a power module for a communication device as an example, defect detection is performed after applying thermal grease to the PCB board. Using 45° side lighting, tiny bubbles were found in the heat dissipation area of ​​the power converter. After processing with a circular feature enhancement filter, five bubbles with diameters between 0.15mm and 0.28mm were successfully identified, and their positions and diameters were recorded. Simultaneously, dual-light source switching was applied to this area, acquiring two sets of images: one from a vertical light source and the other from a 45° side light source. Calculating the difference image revealed uneven thickness transitions at the edge of the heat dissipation area, with the thickness gradient exceeding a preset threshold. Furthermore, high-resolution local magnification imaging was used to inspect the area around the connector, identifying three areas with an area of ​​approximately... Tiny droplet contamination. All defect data are integrated to form a complete defect feature map, including bubble distribution map, thickness unevenness distribution map, and contamination area distribution map.

[0107] In one specific embodiment, the process of executing step S106 may specifically include the following steps:

[0108] (1) Real-time monitoring of production environment parameters through temperature and humidity sensors, recording current factory temperature and humidity values, and generating environmental monitoring data;

[0109] (2) Based on environmental monitoring data, query the temperature-viscosity relationship data and humidity-curing rate relationship data of different types of adhesives to obtain the theoretical performance parameters of the adhesives under the current environment;

[0110] (3) Compare the theoretical performance parameters with the adhesive performance parameters under standard conditions, and calculate the environmental factor compensation coefficient;

[0111] (4) Combine the environmental factor compensation coefficient with the defect type data in the defect feature map, weaken the leveling deviation of silicone in high temperature environment greater than 30 degrees Celsius, weaken the continuity defect of epoxy glue in low temperature environment less than 18 degrees Celsius, and generate the defect evaluation value after environmental compensation.

[0112] (5) To address the differences in the properties of different batches of glue, a batch feature database is established, and the image features of each batch of glue under different environments are recorded. Cross-batch image feature comparison is performed to generate batch compensation coefficients.

[0113] (6) Multiply the environmentally compensated defect evaluation value with the batch compensation coefficient to comprehensively assess the actual defect severity and output a defect judgment result including defect type, location and severity.

[0114] Specifically, temperature and humidity sensors monitor production environment parameters in real time. These sensors are installed at key locations on the production line, no more than 1 meter away from the adhesive application station, ensuring that the measured values ​​accurately reflect the actual application environment of the adhesive. The sensors acquire temperature data with an accuracy of ±0.5℃ and humidity data with an accuracy of ±2%RH, at a sampling frequency of 1 time per minute. The data acquisition module records temperature and humidity values ​​in real time, forming environmental monitoring data including timestamps, temperature values, and humidity values. This environmental monitoring data is transmitted to the detection system via the industrial control network as input parameters for subsequent compensation calculations.

[0115] Based on environmental monitoring data, querying the performance parameter curves of different adhesive types is the foundation for compensation calculations. Each adhesive has specific temperature-viscosity and humidity-curing rate correlation data, which are derived from the technical specifications provided by the adhesive supplier and laboratory test results, and stored in a material parameter database. For thermally conductive silicone grease, its viscosity decreases by approximately 40%-50% for every 10°C increase in temperature; for epoxy resin adhesives, its curing time increases by approximately 2 times for every 10°C decrease in temperature. Through interpolation calculations, the theoretical viscosity, flow parameters, and curing rate of the adhesive under the current environmental temperature and humidity conditions are obtained. These parameters collectively constitute the theoretical performance parameters of the adhesive.

[0116] The theoretical performance parameters are compared with the adhesive performance parameters under standard conditions to calculate the environmental factor compensation coefficient. A standard environment is typically defined as a temperature of 23℃ ± 2℃ and a relative humidity of 50% ± 5%. The adhesive performance parameters measured under this environment are used as the baseline values. The environmental factor compensation coefficient is calculated using the following formula:

[0117]

[0118] in, It is the environmental factor compensation coefficient. This is the theoretical viscosity value of the adhesive under current conditions. This is the viscosity value of the adhesive under standard conditions. This refers to the flowability parameter of the adhesive under current conditions. This is the flowability parameter of the adhesive under standard conditions. This calculation takes into account the combined effects of temperature and humidity changes on the adhesive's viscosity and flowability, forming a comprehensive compensation factor that reflects environmental differences.

[0119] By combining environmental factor compensation coefficients with defect type data from defect feature maps, defect assessments under different environmental conditions are adjusted. In high-temperature environments (above 30℃), silicone leveling properties are enhanced, resulting in more pronounced diffusion than at normal temperatures; therefore, defects related to leveling deviations are mitigated. In low-temperature environments (below 18℃), epoxy adhesive viscosity increases, making it prone to breakage and discontinuity; therefore, continuous defects are mitigated. This mitigation is achieved by adjusting defect assessment weights, and the severity scores of various defects are corrected based on the environmental factor compensation coefficients to obtain the environmentally compensated defect evaluation value.

[0120] To address the differences in adhesive properties between batches, establishing a batch feature database is crucial for resolving batch-to-batch variations. Each batch of adhesive exhibits slight differences in composition and properties, which manifest differently under varying environmental conditions. The batch feature database records the image feature parameters of each batch of adhesive under typical environmental conditions, including color, texture, and reflectivity. When a new batch of adhesive is put into use, it is first tested using standard samples, collecting its image features under different environments. These features are then compared with historical batches in the database to calculate the feature difference and generate a batch compensation coefficient. The batch compensation coefficient reflects the degree of deviation between the current batch and the standard batch and is used to adjust defect assessment criteria.

[0121] The environmentally compensated defect evaluation value is multiplied by the batch compensation coefficient to obtain the final defect assessment value, which comprehensively considers environmental factors and batch differences. This final assessment value is used to determine the actual severity of the defect. The final defect judgment result includes three key pieces of information: defect type (e.g., bubbles, uneven thickness, contamination), location information (pixel-level coordinates), and severity (classified as minor, moderate, and severe). This comprehensive evaluation process ensures the consistency and reliability of defect judgment results under different environmental conditions and material batches.

[0122] Taking the production of a power module for a certain type of communication equipment as an example, adhesive application defects were detected under winter workshop conditions of 16℃ and 35%RH. The epoxy resin used had a viscosity of 15000 cP and a flowability parameter of 0.75 under standard conditions (23℃ / 50%RH). However, under the current low temperature and low humidity conditions, the theoretical viscosity obtained from the material parameter curve was 26000 cP, and the flowability parameter was 0.45. Based on this, the environmental factor compensation coefficient was calculated to be 1.73, indicating that the adhesive flowability was significantly reduced under the current environment. During the detection process, the defect feature map showed three continuous defects in the epoxy adhesive on the PCB board. However, considering the high viscosity of the adhesive caused by the low temperature environment, the severity score of these continuous defects was weakened by about 40%. At the same time, a query from the batch feature database revealed that the epoxy adhesive currently used was a new batch, and its image features under low temperature conditions differed from those of the standard batch by 0.85. Based on this, a batch compensation coefficient of 0.92 was generated. Multiplying the environmentally compensated defect evaluation value by the batch compensation coefficient, the final defect judgment result shows that 2 out of 3 continuous defects are judged as "minor" and 1 is judged as "moderate".

[0123] The above describes the glue application defect detection method based on image processing in the embodiments of this application. The following describes the glue application defect detection system based on image processing in the embodiments of this application. Please refer to [link to relevant documentation]. Figure 2 One embodiment of the glue application defect detection system based on image processing in this application includes:

[0124] The acquisition module is used to illuminate the glue application area of ​​the PCB board in the electronic manufacturing production line using a narrow-band blue LED light source, acquire glue application images of the PCB board, and obtain raw glue image data.

[0125] The recognition module is used to identify different colors of glue based on the original glue image data by selecting the adaptive color separation channel, and obtain glue area feature data.

[0126] The analysis module is used to perform trajectory analysis on the glue distribution in a specific area of ​​the product based on the glue area feature data and through three-dimensional glue path modeling, and to establish a three-dimensional glue model that includes planar position and thickness distribution.

[0127] The judgment module is used to differentiate the glue application quality of key functional areas, secondary functional areas and non-functional areas of the PCB board based on the glue three-dimensional model and through multi-level tolerance evaluation, and generate regional defect data.

[0128] The enhancement module is used to perform feature enhancement analysis on microbubbles, uneven thin-layer coverage, and glue contamination based on the regional defect data, through specific angle side lighting and dual light source switching, to form a defect feature map;

[0129] The adjustment module is used to dynamically adjust the parameters of the adhesive performance differences caused by changes in the production environment based on the defect feature map and through a temperature and humidity compensation algorithm, and output the defect judgment result.

[0130] Through the collaborative efforts of the aforementioned components, a narrow-band blue LED light source is used to precisely illuminate the adhesive application area on the PCB board in the electronic manufacturing production line, achieving optimal contrast between different colored adhesives and the PCB board background, significantly improving the quality of the original image data. Then, an adaptive color separation channel selection technology is employed to automatically select the optimal recognition channel based on the spectral characteristics of different colored adhesives, overcoming the limitation of traditional single-channel methods that cannot adapt to multiple adhesive types. The 3D adhesive path modeling process fully considers the three-dimensional morphological characteristics of the adhesive, analyzing not only planar distribution but also thickness information, extending detection from two-dimensional to three-dimensional, better meeting practical application needs. A multi-level tolerance evaluation mechanism sets differentiated detection standards for different functional areas on the PCB board, avoiding excessive rejection or missed detection caused by a one-size-fits-all approach. Specific-angle side lighting and dual-light source switching technology significantly improve the detection capability of difficult-to-detect defects such as micro-bubbles, uneven thin-layer coverage, and micro-adhesive droplet contamination, achieving a detection accuracy on the order of 0.1mm. The temperature and humidity compensation algorithm innovatively incorporates environmental factors, automatically adjusting judgment parameters and solving the detection fluctuation problem caused by environmental changes. This invention organically integrates artificial intelligence algorithms into the glue defect detection process. Particularly in the circular feature enhancement filtering, multi-level tolerance assessment, and temperature and humidity compensation stages, the algorithmic features significantly improve the system's adaptability and robustness, enabling the detection process to adaptively handle complex scenarios under different working conditions. The comprehensive application of these technical features not only greatly improves the accuracy and consistency of glue application defect detection, but more importantly, it can adapt to various application scenarios in the electronics manufacturing industry, including key application areas such as power device heat dissipation, chip sealing, and waterproofing and dustproofing, greatly improving product quality and reliability. Furthermore, by establishing a batch feature database, this invention solves the problem of fluctuations in detection standards caused by different batches of glue, providing a stable and consistent quality control capability for the production process and improving detection efficiency.

[0131] Reference Figure 3 This invention also provides a computer device, which can be a server, and its internal structure can be as follows: Figure 3 As shown, the computer device includes a processor, memory, display screen, input device, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database stores the data corresponding to this embodiment. The network interface is used to communicate with external terminals via a network connection. When the computer program is executed by the processor, it implements the above-described method.

[0132] Those skilled in the art will understand that Figure 3 The structures shown are merely block diagrams of some structures related to the present invention and do not constitute a limitation on the computer devices on which the present invention is applied.

[0133] An embodiment of the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method. It is understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0134] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the present invention and embodiments can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual-rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM, etc.

[0135] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0136] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

Claims

1. A method for detecting glue application defects based on image processing, characterized in that, The image processing-based glue application defect detection method includes: The adhesive application area of ​​the PCB board in the electronic manufacturing production line is illuminated by a narrow-band blue LED light source, and the adhesive application image of the PCB board is collected to obtain the original adhesive image data. Based on the original glue image data, different colored glues are identified and processed by adaptive color separation channel selection to obtain glue area feature data. Based on the adhesive area feature data, the adhesive distribution in a specific area of ​​the product is analyzed by three-dimensional adhesive path modeling to establish a three-dimensional adhesive model that includes planar position and thickness distribution. Based on the three-dimensional model of the adhesive, the adhesive application quality of key functional areas, secondary functional areas and non-functional areas of the PCB board is differentiated by multi-level tolerance assessment, and regional defect data is generated. Based on the regional defect data, feature enhancement analysis is performed on microbubbles, uneven thin-layer coverage, and glue contamination by using side lighting at a specific angle and switching between dual light sources to form a defect feature map. Based on the defect feature map, the temperature and humidity compensation algorithm is used to dynamically adjust the parameters of the adhesive performance differences caused by changes in the production environment, and output the defect judgment result.

2. The method for detecting glue application defects based on image processing according to claim 1, characterized in that, The process involves illuminating the adhesive application area of ​​a PCB board in the electronic manufacturing production line using a narrow-band blue LED light source, acquiring images of the adhesive application on the PCB board, and obtaining raw adhesive image data, including: The narrowband blue LED light source is set to a wavelength range of 470nm-480nm, and the narrowband blue LED light source is arranged above the PCB board surface at an incident angle of 30° to avoid glare interference from the PCB surface. High-definition imaging of the PCB board is performed at a fixed position 300mm behind the adhesive coating station of the SMT production line, covering a maximum area of ​​350mm×280mm of the PCB board; Set the camera shutter speed to 1 / 1000 second to ensure clear images are captured while the production line is in motion; By detecting the PCB board's arrival signal using a photoelectric sensor, the image acquisition system is precisely triggered to capture a full-coverage image of the area on the PCB board where adhesive has been applied. The preset lighting parameters are automatically switched according to different product models to enhance the directional lighting of the glue application area on the PCB board. The high-definition image data of the PCB board glue application area is collected, encoded and stored to generate the original glue image data.

3. The method for detecting glue application defects based on image processing according to claim 1, characterized in that, The process involves identifying different colored adhesives based on the original adhesive image data using adaptive color separation channel selection to obtain adhesive region feature data, including: A standard grayscale reference block is preset on the edge of the PCB board. The current illumination compensation coefficient is calculated by comparing the difference ratio between the actual brightness value of the reference block and the standard value. The original glue image data is multiplied by the illumination compensation coefficient to generate a standardized illumination image, and the standardized illumination image is decomposed into single-channel images with three RGB channels. By calculating the signal-to-noise ratio of the single-channel image, the single channel with the highest contrast is determined as the glue recognition channel. The PCB silkscreen and copper foil background interference areas in the glue recognition channel are masked to generate a background-removed image; For transparent or semi-transparent silicone areas in the background removal image, reflections are eliminated by adjusting the parameters of the polarizing filter to generate a boundary enhancement image; The boundary enhancement image is analyzed for shape regularity and edge transition characteristics to extract the position coordinates, area values, and boundary data of the glue application area, forming the feature data of the glue area.

4. The method for detecting glue application defects based on image processing according to claim 1, characterized in that, Based on the adhesive region feature data, the process involves performing trajectory analysis on the adhesive distribution in a specific area of ​​the product through three-dimensional adhesive path modeling, and establishing a three-dimensional adhesive model that includes planar position and thickness distribution, including: Import CAD drawings of PCB products from the production design database, extract theoretical glue path information, and generate basic glue path outline diagram; Spatial registration is performed between the location coordinates, area values, and boundary data of the glue area feature data and the basic glue path outline map to form an actual glue path distribution map. The actual adhesive path distribution map is classified and labeled into three types of adhesive application areas: mandatory application area, prohibited application area, and optional application area. Establish characteristic parameter tables for different types of adhesives, record the shrinkage parameters of epoxy resin adhesives and the leveling parameters of silicone adhesives, and construct a library of physicochemical properties of adhesives; Based on the thermal paste distribution conditions in the installation of server heat sinks, a thickness distribution curve is added to the actual adhesive path distribution diagram, wherein the thermal paste distribution conditions are thicker in the middle and thinner at the edges; The actual adhesive path distribution map is combined with the thickness distribution curve to form a three-dimensional model of the adhesive that includes planar position and thickness information.

5. The method for detecting glue application defects based on image processing according to claim 1, characterized in that, Based on the three-dimensional model of the adhesive, the adhesive application quality of key functional areas, secondary functional areas, and non-functional areas of the PCB board is differentiated through multi-level tolerance assessment, generating regional defect data, including: The PCB board is divided into functional areas, namely the critical functional area of ​​the power device heat dissipation surface and waterproof sealing line, the secondary functional area of ​​the fixing bracket and auxiliary heat dissipation area, and the non-functional area of ​​the marking area. A first-level tolerance standard is established for the key functional areas, setting a judgment threshold of glue coverage of not less than 95% and thickness error not exceeding ±10%. A second-level tolerance standard is established for the secondary functional area, setting a judgment threshold of glue coverage of not less than 85% and thickness error not exceeding ±20%. A third-level tolerance standard is established for the non-functional area, which mainly detects glue overflow pollution and sets a judgment threshold that the overflow area does not exceed 0.5 square millimeters; Based on the glue distribution data in the glue 3D model, the glue application deviation value of each functional area is calculated by comparing the first-level tolerance standard, the second-level tolerance standard, and the third-level tolerance standard. The glue application deviation value of each functional area is compared with the tolerance standard of the corresponding functional area to generate the sub-regional defect data, which includes glue missing, glue overflow and glue thickness abnormality.

6. The method for detecting glue application defects based on image processing according to claim 1, characterized in that, The process involves performing feature enhancement analysis on microbubbles, uneven thin-layer coverage, and adhesive contamination based on the regional defect data, using specific angle side lighting and dual-source switching, to form a defect feature map, including: For the microbubble regions in the segmented defect data, a specific angle of side lighting is set, with the light source angle being 30°-60°, to enhance the reflective properties of the bubble edges and generate an enhanced bubble image; The bubble enhancement image is then processed by applying a circular feature enhancement filter to highlight the outline of tiny bubbles with a diameter of less than 0.3 mm, thereby obtaining bubble feature data. The thin-layer uneven coverage area in the regional defect data is illuminated by switching between two light sources, using a 0° vertical light source and a 45° side light source in sequence to obtain two sets of illumination images; Calculate the difference map between the two sets of illumination images, extract areas with significant differences in reflectivity, and form thickness unevenness feature data; For the glue-contaminated areas in the segmented defect data, high-resolution local magnification imaging and texture analysis are used to identify tiny glue droplets with a size as small as 0.1mm × 0.1mm, generating feature data of the contaminated areas; The bubble feature data, thickness unevenness feature data, and contaminated area feature data are integrated to form the defect feature map, which includes defect type, location information, and severity.

7. The method for detecting glue application defects based on image processing according to claim 1, characterized in that, Based on the defect feature map, the system dynamically adjusts the parameters of the adhesive performance differences caused by changes in the production environment using a temperature and humidity compensation algorithm, and outputs the defect judgment result, including: Real-time monitoring of production environment parameters is achieved through temperature and humidity sensors, recording current factory temperature and humidity values, and generating environmental monitoring data. Based on the environmental monitoring data, query the temperature-viscosity relationship data and humidity-curing rate relationship data of different types of adhesives to obtain the theoretical performance parameters of the adhesives under the current environment; The theoretical performance parameters are compared with the adhesive performance parameters under standard conditions, and the environmental factor compensation coefficient is calculated. By combining the environmental factor compensation coefficient with the defect type data in the defect feature map, the leveling deviation of silicone in high temperature environment greater than 30 degrees Celsius is weakened, and the continuity defect of epoxy resin in low temperature environment less than 18 degrees Celsius is weakened, thus generating a defect evaluation value after environmental compensation. To address the differences in adhesive properties between different batches, a batch feature database is established to record the image features of each batch of adhesive under different environments. Cross-batch image feature comparisons are then performed to generate batch compensation coefficients. The environmentally compensated defect evaluation value is multiplied by the batch compensation coefficient to comprehensively assess the actual defect severity and output the defect judgment result, which includes the defect type, location, and severity.

8. An image processing-based glue application defect detection system, used to implement the image processing-based glue application defect detection method as described in any one of claims 1-7, characterized in that, The image processing-based glue application defect detection system includes: The acquisition module is used to illuminate the glue application area of ​​the PCB board in the electronic manufacturing production line using a narrow-band blue LED light source, acquire glue application images of the PCB board, and obtain raw glue image data. The recognition module is used to identify different colors of glue based on the original glue image data by selecting the adaptive color separation channel, and obtain glue area feature data. The analysis module is used to perform trajectory analysis on the glue distribution in a specific area of ​​the product based on the glue area feature data and through three-dimensional glue path modeling, and to establish a three-dimensional glue model that includes planar position and thickness distribution. The judgment module is used to differentiate the glue application quality of key functional areas, secondary functional areas and non-functional areas of the PCB board based on the glue three-dimensional model and through multi-level tolerance evaluation, and generate regional defect data. The enhancement module is used to perform feature enhancement analysis on microbubbles, uneven thin-layer coverage, and glue contamination based on the regional defect data, through specific angle side lighting and dual light source switching, to form a defect feature map; The adjustment module is used to dynamically adjust the parameters of the adhesive performance differences caused by changes in the production environment based on the defect feature map and through a temperature and humidity compensation algorithm, and output the defect judgment result.

9. A computer device, characterized in that, The method includes a memory and a processor, the memory storing a computer program that can run on the processor, characterized in that, when the processor executes the computer program, it implements the image processing-based glue application defect detection method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, the computer program, when executed by a processor, causing the processor to perform the image processing-based glue application defect detection method as described in any one of claims 1 to 7.

Images