| During the manufacturing of printed circuit boards, multiple layers of materials are joined together and in order to ensure electrical connectivity, the layers of material must be aligned. The alignment is known as registration, signifying the targets based upon fiducial markers of where the individual layers should lie. Failure of the alignment of these layers causes defects, which creates a loss of electrical signal.;This research utilized X-ray measurements of an inspection process to correctly classify yield status of the printed circuit board prior to the known status of the board itself. The developed classification models include multiple linear regression, stepwise multiple regression, and artificial neural networks. Dimension reduction techniques such as principal component analysis and cluster analysis had also been examined. With limited number of defective samples, a cloning method was used to decrease the disproportion bias error of the data and increase the number of defective samples. |
