MLPCA based logistical regression analysis for pattern clustering in manufacturing processes

Abstract

This article proposes a method of improving pattern clustering accuracy combined with a logistical regression model for manufacturing processes with binary inspection outputs. A latent variable model was incorporated into the classical logistical regression model, involving the use of Maximum Likelihood Principal Component Analysis (MLPCA) to identify the underlying variation sources governing the behaviors of the high dimensional measurable variables. The highly correlated continuous measurable predictors are projected onto a lower dimensional latent space, followed by a more precise pattern clustering algorithm to the inspected manufacturing products for in-line process monitoring. The example of the visual inspects from semiconductor manufacturing processes is shown that this new pattern clustering algorithm could help identify the root causes of the variations. © Springer-Verlag 2003.