Process oriented basis estimation in presence of non-orthogonal basis elements.

摘要

Process Oriented Basis Representations (POBREP) is a multivariate Statistical Process Control (SPC) procedure with diagnosis capabilities developed by Barton and Gonzalez-Barreto (1996). Although this methodology is effective when orthogonal process-oriented basis (POB) is presented, it is diagnosis capabilities are at risk when the POB is not orthogonal. This research compared several methods to solve non-orthogonal POB's problem. Six scenarios with different Variance Inflation Factor (VIF) severity were created using the stencil printing process. Coefficients were estimated using five methods: Ordinary Least Square (OLS), Independent Subsets (IS), Simple Regression (SR), Ridge Regression (RR) and Constrained Solution Space (CSS). These methods were compared in terms of the lower Square Error (SE) and higher number of times the coefficient is between a confidence interval (Count). There were two comparable groups of results: (1) CSS and RR methods with lowest SE and highest Count and (2) OLS, IS and SR with higher SE and lower Count. The best method estimate POBREP coefficient in presence of non non-orthogonal basis elements is Constraint Space Solution.