CN-121981942-A - Circuit board welding spot detection method based on machine vision

CN121981942ACN 121981942 ACN121981942 ACN 121981942ACN-121981942-A

Abstract

The application provides a machine vision-based circuit board welding spot detection method, which comprises the steps of carrying out multi-angle scanning on a welding spot area through a high-resolution camera to generate a three-dimensional welding spot image, extracting contour lines of a welding spot and welding spot contact interface from the three-dimensional welding spot image, calculating contour curvature distribution data to obtain a welding spot wetting angle, obtaining an accurate wetting angle value through iterative optimization adjustment calculation parameters, carrying out pixel density distribution analysis on the welding spot area according to the accurate wetting angle value, identifying a density abnormal area, merging adjacent abnormal areas into a potential defect area, carrying out feature extraction and pattern matching on the potential defect area by utilizing a convolutional neural network, identifying defect types, integrating position, type and severity information of all defect areas, and generating a welding spot quality comprehensive evaluation report by combining the wetting angle data. The application solves the problem that the prior art is difficult to fuse the quantitative evaluation of the wetting angle of the welding spot with the defect identification depth, so that the detection is inaccurate.

Inventors

ZHANG FATIAN
ZHANG SHUPING
Zhang Wanmo
ZHANG ZIMING
ZHONG MINGHUA
ZHOU FUPING
ZHOU TAI

Assignees

鹤山市泰利诺电子有限公司

Dates

Publication Date: 20260505
Application Date: 20251211

Claims (8)

1. The machine vision-based circuit board welding spot detection method is characterized by comprising the following steps of: performing multi-angle scanning on the welding spot area through a high-resolution camera, collecting sequence images and generating three-dimensional welding spot images based on feature matching and a three-dimensional reconstruction algorithm; extracting contour lines of a welding spot and welding pad contact interface from the three-dimensional welding spot image, and calculating contour curvature distribution data; Acquiring a welding spot wetting angle based on the profile curvature distribution data, and when the wetting angle is smaller than a preset threshold value, obtaining an accurate wetting angle value by iteratively optimizing and adjusting calculation parameters; Carrying out pixel density distribution analysis on the welding area according to the accurate wetting angle value, identifying a density abnormal area, merging adjacent abnormal areas through a clustering algorithm, and marking the adjacent abnormal areas as potential defect areas; performing feature extraction and pattern matching on the potential defect area by using a predetermined convolutional neural network, and identifying the defect type; And integrating the position, type and severity information of all the defect areas, and generating a welding spot quality comprehensive evaluation report by combining the wetting angle data.
2. The machine vision based circuit board solder joint detection method of claim 1, wherein the three-dimensional solder joint image comprises: comprehensively scanning a welding area by adopting a high-resolution camera to acquire multi-angle view data, so as to construct initial three-dimensional view information and form a preliminary acquisition data set; based on the preliminary acquisition data set, integrating multi-angle information by using an image fusion technology to generate a three-dimensional welding spot image; Performing preset denoising processing on the three-dimensional welding spot image to obtain a corrected three-dimensional welding spot image; if the corrected three-dimensional welding spot image has local data missing, the corrected three-dimensional welding spot image is complemented by an interpolation technology to obtain a complete three-dimensional welding spot image; Extracting the surface information of a welding spot and evaluating the flatness data of the welding spot by carrying out layering analysis on the complete image; Based on the flatness data, classifying the surface quality of the welding spots by adopting a support vector machine algorithm, and determining quality grade information of the welding spots; and if the quality grade is lower than a preset threshold, marking the corresponding position of the welding spot area, and generating marked three-dimensional welding spot image data.
3. The method of claim 1, further comprising removing light and angle interference noise from the three-dimensional solder joint image by adaptive filtering, and enhancing edge contrast to obtain an enhanced solder joint image.
4. A machine vision based circuit board solder joint detection method according to claim 3, wherein said calculating profile curvature distribution data comprises: the contour feature data obtained from the reinforced welding spot image are used for carrying out region division on contour lines of a welding spot interface by adopting a layering segmentation technology, so that layered contour region data are obtained; According to the layered contour region data, a local contrast analysis method is applied to calculate the curvature change of each region section by section to determine contour curvature distribution data, wherein the method further comprises the following steps: Aiming at the profile curvature distribution data, if the curvature change of a certain area is found to exceed a preset threshold range, marking the area through an abnormal marking technology, and obtaining marked area data; classifying the wetting form of the identification area by adopting a support vector machine algorithm according to the marked area data, and judging classified form grade data; Aiming at the classified form grade data, matching the classified form grade data with a preset wetting form standard library through a data comparison technology to obtain matched form consistency data; If the consistency is lower than a preset threshold value in the matched form consistency data, storing the contour line information of the area into an abnormal file database through a data recording technology, and determining an abnormal file record; and according to the abnormal file records, classifying and storing the abnormal file records into a preset welding spot morphology analysis library by adopting an automatic archiving technology, and obtaining structured morphology analysis data.
5. The machine vision based circuit board solder joint detection method of claim 1, wherein the precise wetting angle value comprises: the tangent line data and the normal intersection point data of the contact interface are extracted from the profile curvature distribution data, the included angle value of the tangent line data and the normal intersection point data is determined through a geometric calculation method, and preliminary included angle data are obtained; if the preliminary included angle data of a certain area is smaller than a preset threshold value, judging that the area is insufficient in wetting, and marking the area through a marking technology to obtain marked area information; correcting the calculated parameters by adopting an iterative adjustment method according to the marked region information, optimizing parameter configuration in a successive approximation mode, and determining adjusted parameter data; and (3) recalculating the wetting angle value through the adjusted parameter data, and calibrating the angle value by adopting a standard geometric analysis tool to obtain accurate wetting angle data.
6. The machine vision based circuit board solder joint detection method of claim 1, wherein the potential defect area comprises: The pixel distribution in the target area is initially scanned by combining the wetting angle data with the contour curvature distribution data, and the area is divided into a plurality of subunits by adopting a grid division technology to obtain pixel density data of each subunit; calculating density variation conditions aiming at pixel density data of each subunit, and if the density variation of a certain subunit exceeds a preset threshold value, marking the subunit as an abnormal subunit, and determining an abnormal subunit set; And extracting the spatial correlation between adjacent subunits according to the abnormal subunit set, and merging the adjacent abnormal subunits into a potential defect area by adopting a cluster analysis method.
7. The machine vision based circuit board solder joint detection method of claim 1, wherein the identifying the defect type comprises: Processing the labeling data through a preset convolutional neural network, and executing feature extraction operation on the image information in the defect area to obtain an extracted feature data set; according to the characteristic data set, a mode comparison technology is adopted to match with a preset mode, if the matching result accords with the characteristics of the air hole mode, the corresponding area is marked as a high risk area, and a preliminary classification label is determined; Aiming at the preliminary classification labels, acquiring image detail data in a high risk area, adopting a pixel analysis tool to conduct subdivision judgment on defect types, and if crack features exist in the detail data, updating the classification labels into crack types to obtain subdivided defect classification information; Extracting position coordinate data of the high risk area according to the subdivided defect classification information, and associating the position coordinate with the defect type by adopting a data mapping technology to obtain an associated position classification record; acquiring image layering data of a corresponding area according to the associated position classification record, confirming the distribution range of the defect types through a layering comparison tool, and determining boundary data of the distribution range; Integrating the defect type, the position coordinates and the distribution range information into a structured record by adopting a data storage technology through boundary data of the distribution range to obtain final classified archival data; And storing the classification information and the position data of all the high risk areas into a defect database by adopting an automatic filing tool according to the final classification file data to obtain a complete defect management record.
8. The machine vision based circuit board solder joint detection method of claim 1, wherein generating a solder joint quality integrated assessment report comprises: acquiring position information and severity data of a high-risk region from a classified defect result, and carrying out structural processing on the information through a data sorting tool to obtain a sorted region risk data set; Aiming at the sorted regional risk data sets, carrying out item-by-item comparison by adopting a data comparison method and a preset threshold value, if the severity data of a certain region exceeds the preset threshold value, marking the region as an abnormal region, and determining an abnormal region list; acquiring welding spot wetting image data of a corresponding region according to the abnormal region list, and extracting details of the image data by an image processing tool to obtain a detail feature set of welding spot wetting; Aiming at the detail feature set of welding spot wetting, classifying the feature set by adopting a convolutional neural network, judging whether the phenomenon of insufficient wetting exists, and obtaining a classified wetting state record; Acquiring associated data of the abnormal region and the wetting state according to the classified wetting state record, integrating the associated data with the position information through a data mapping tool, and determining a comprehensive quality evaluation data set; For the comprehensive quality evaluation data set, carrying out overall judgment on the wetting states of all abnormal areas through an automatic analysis tool, and if the overall quality does not reach a preset threshold value, marking the abnormal state as an unqualified state to obtain final quality judgment output; and integrating the position information, the severity data and the judging result of all the high-risk areas into a structured file through a data storage tool according to the final quality judging output, and obtaining a complete circuit board reliability analysis record.

Description

Circuit board welding spot detection method based on machine vision Technical Field The invention relates to the technical field of circuit board welding spot detection, in particular to a machine vision-based circuit board welding spot detection method. Background In the electronics industry, solder joint quality is directly related to the reliability of the circuit board and even the entire electronic product. Automatic optical detection systems are commonly used in the industry at present, and appearance defects of welding spots, such as bridging, tin-free, cold joint, offset and the like, are mainly identified through a two-dimensional imaging mode. However, such methods have significant limitations in that they are based on apparent characteristics such as morphology, color, or brightness, and are difficult to access to the essential process parameters of the weld quality. The wetting angle directly reflects the integrity and strength of the metal bond between the solder and the pad, which is the core basis for evaluating solder reliability. In conventional visual inspection, measuring the wetting angle typically requires destructive sectioning of the sample and is accomplished in a laboratory environment with the aid of a microscope or scanning electron microscope, a process that is time consuming and not applicable to the production of a full inspection. In the prior art, the defect detection and the process quality evaluation are divided into two independent links, however, the defect detection system can only judge whether the abnormality exists or not, but cannot judge why the abnormality exists, and the process evaluation is often dependent on spot check and experience inference due to lack of online data support. The quality control closed loop is incomplete due to the fracture, the problem root analysis is difficult, and the process optimization lacks accurate basis. Disclosure of Invention The invention provides a machine vision-based circuit board welding spot detection method, and aims to solve the problem that in the prior art, quantitative evaluation of a welding spot wetting angle is difficult to fuse with defect identification depth, so that detection is inaccurate. In order to achieve the above purpose, the invention provides a machine vision-based circuit board welding spot detection method, which comprises the following steps: performing multi-angle scanning on the welding spot area through a high-resolution camera, collecting sequence images and generating three-dimensional welding spot images based on feature matching and a three-dimensional reconstruction algorithm; extracting contour lines of a welding spot and welding pad contact interface from the three-dimensional welding spot image, and calculating contour curvature distribution data; Acquiring a welding spot wetting angle based on the profile curvature distribution data, and when the wetting angle is smaller than a preset threshold value, obtaining an accurate wetting angle value by iteratively optimizing and adjusting calculation parameters; Carrying out pixel density distribution analysis on the welding area according to the accurate wetting angle value, identifying a density abnormal area, merging adjacent abnormal areas through a clustering algorithm, and marking the adjacent abnormal areas as potential defect areas; performing feature extraction and pattern matching on the potential defect area by using a predetermined convolutional neural network, and identifying the defect type; And integrating the position, type and severity information of all the defect areas, and generating a welding spot quality comprehensive evaluation report by combining the wetting angle data. Further, the three-dimensional solder joint image includes: comprehensively scanning a welding area by adopting a high-resolution camera to acquire multi-angle view data, so as to construct initial three-dimensional view information and form a preliminary acquisition data set; based on the preliminary acquisition data set, integrating multi-angle information by using an image fusion technology to generate a three-dimensional welding spot image; Performing preset denoising processing on the three-dimensional welding spot image to obtain a corrected three-dimensional welding spot image; if the corrected three-dimensional welding spot image has local data missing, the corrected three-dimensional welding spot image is complemented by an interpolation technology to obtain a complete three-dimensional welding spot image; Extracting the surface information of a welding spot and evaluating the flatness data of the welding spot by carrying out layering analysis on the complete image; Based on the flatness data, classifying the surface quality of the welding spots by adopting a support vector machine algorithm, and determining quality grade information of the welding spots; and if the quality grade is lower than a preset threshold, marking the corresponding position