In this thesis, a new algorithm for PCB inspection using x-ray was developed. Using X-ray, inspection of not only visible but invisible defect is possible, because x-ray has transmission characteristics. Algorithm has three inspection steps. At the first step, solder state is determined by backpropagation neural network by classifying input patterns which is the slice areas with different heights of solder joint. Slice areas are extracted with high speed utilizing ILUT and histogram operation. At the second step, defects as bridge or solder ball are detected by counting the number of object regions in the window for bridge/solder ball inspection. This step utilizes quad tree segmentation to separate objects from background. At the last step, IC rotation is inspected by calculating the solpe of the regression line of solder points which are captured in a view field of the camera.
The computer simulation using real x-ray image of the solder joints shows high noise immunity and high speed inspection which is not affected whether the leads are visible or not.