The Study On Low-level Vision Algorithms Used In PCB Automatic Optical Inspection Systems

With the development of manufacturing technology, the trend of miniaturization and integration in military and industrial products is more and more evident. This required higher than ever with the quality of manufacturing process of printed circuit board(PCB).PCB automatic optical inspection(AOI) devices are important to guarantee these process of PCB, further more, the technology of image processing and analysis play a key role in these equipments. Many researches have been focused on this field in some developed countries or regions such as America, Japan, Israel and Taiwan, most equipments used in our country are coming from these countries and regions. So the study of the core technology in this field such as image processing and analysis is very important in the development of our devices.As the fast development of Chinese manufacturing industry in recent years, developing our high-level devices is very important, in the procedure of industrial structure adjustment and promotion. However, the localization is still of a low level, since we started late and the foundation is poor. To shorten the distance with technological leaders, the thorough and systematic research work on the key techniques of the PCB AOI are necessary.Combining with practical projects background, some low-level algorithms in the AOI are discussed in this dissertation, and some new approaches are proposed.1. A color image de-noising algorithms has been designed for PCB AOI devices.The stability of color inspection in PCB AOI is still a challenge problem in devices coming from western world. The texture pattern and the unstable color distribution in pad area make it difficult to find all color defects in these region consistently.In the study of color defects detection in PCB final visual inspection devices, the noise level and color distribution have been analyzed carefully. In order to guarantee the stability of the final effect and refine the raw data, a method combined the technology of Bilateral Filter and ***color space has been used. This method suppressed noise effectively, in the same time, high quality pre-processed data are generated which will be given to the classifier in the following detection procedure.2. A automatic compensation method for sub-pixel edge detection has been designed for bi-telecentric measure system.The bi-telecentric system provides the possibility of real sub-pixel accuracy in measurement. However, after the analysis of traditional sub-pixel edge detection algorithms,the problem of sub-pixel bias in position has been found. An automatic compensation system has been designed, which compose of image analysis technology, statistical learning skill and radial basis function neural network.With this system, the position bias in traditional sub-pixel edge detection algorithms can be compensated effectively in an automatical way. In the final products, 5 accuracy has been achieved in the resolution of 45 / and 1 accuracy in the resolution at 18.7 /.3. The calibration system for area-scan and line-scan AOI system has been studied.In area-scan system, the accuracy of calibration has been analyzed carefully. The result of the analysis can guide the calibration procedure effectively.In line-scan system,a real-time visualization of the precision light machine tuned technology was proposed. The line array direction was used to establish spatial coordinates, the parameters of the space coordinates of the degrees of freedom was calculated by image acquisition analysis. The degrees of freedom parameter characterize the relative position of the camera coordinate system, so the camera position can be adjusted in real time by adjusting the parameters of the degrees of freedom.Theoretical analysis and experimental results show that: the optical resolution can achieve 2.7 through the method, and the analysis and tuning can be synchronized.4. A new two-dimensional image mosaic method which based on sub-pixel grating positioning is proposed.Focused on the problems that it’s difficult to obtain high precision mosaic image of object which has duplicate texture in the automatic visual inspection system, this method using grating to record positions of object and calibration plate to accurately calibrate x and y axis’ s angles between platform and camera, building a 2d stitching model based on global Cartesian coordinate system to complete image stitching.Experimental results show that the method’s calibration repeatability reaches 10-4 radians order of magnitude, average physical error is 8, stitching precision increased nearly 8 times than not correction, which is 2 times better than the method that has been reported.The method has been successfully used in the online automatic inspection equipment for ceramic substrate of navigation satellite’s receiving antenna circuit. This research result can be widely applied in machine vision automated inspection fields that require sub-pixel level detection precision.5. A search method of AOI system based on optimized K-D tree is proposed.The execution time for the method to build up the K-D tree and perform the whole searching is proportional to ( log ).The method has been successfully used in PCB hole AOI system, which is the first Hole AOI device in Chinese mainland.From the aspects mentioned above, some important problems of low-level algorithms in the PCB AOI application are resolved. And the transformation of PCB AOI equipments from theory to application could be strongly accelerated by our research results.