Prediction of Laser Cutting Qualities for BGA Strip by Artificial Neural Networks

This paper reports a new laser cutting technology for cut ting a BGA (ball grid array) strip. Due to the cutting path including two different materials, i.e. epoxy and copper substrate, so we adopt two cutting processes for epoxy and copper substrate, respectively. Then we utilize artificial neural networks (ANN) to build two predictive models including epoxy and substrate for laser cutting qualities of a strip BGA. Two ANN models use the back-propagation (BP) with Levenberg-Marquardt (LM) algorithm. From the experimental results, the average predicted errors of epoxy for training and testing processes are 0.496% and 1.786%, respectively, and the errors for substrate are 0.797% and 1.532%, respectively. The results show that two ANN models have the predictive ability to estimate three laser cutting qualities for BGA strip accurately. The ANN applied to this paper is very successfully and may give guides in the predictions of cutting BGA strip and is expected to be useful for laser applications in other industry fields.