STATISTICAL NEURAL NETWORK MODELING FOR STENCIL PRINTING
Author: Roop L. Mahajan Company: University of Colorado Date Published: 9/10/1996
Surface Mount International
Abstract: Stencil printing is the first and perhaps the most critical step affecting the quality and yield of a fine pitch surface mount assembly. This process must be optimized for higher assembly yield. In this paper, a statistical design of experiment (DOE)-based neural network approach is presented that serves to economically model the process. It is shown that a neural network model based on a Taguchi L27 orthogonal array does an excellent job of determining the optimum settings for minimum solder paste height variation. The effect of optimum settings on the fatigue life of solder joints is presented. Finally, techniques are proposed to transfer a neural network process model from one machine to another without the need for extensive experimental data.