Multiple Response Process Optimization Using Process Capability

• Activity Number: 556
• Type: Roundtables
• Date/Time: Wednesday, August 7, 2013 : 12:30 PM to 1:50 PM
• Sponsor: Quality and Productivity Section
• Author(s): John Peterson*+
• Companies: GlaxoSmithKline
• Keywords: Response surface methodology ; Bayesian ; Reliability ; ICH Q8 Design Space ; Predictive inference

Abstract

Quality improvement can be defined as "reduction of variation about a target." For multiple response process optimization, this can be envisaged as shifting and shrinking of the response distribution about a vector of target values. We also can think of this as optimizing the joint probability that the process will produce responses that meet all specifications. In fact, multivariate process capability can be quantified simply as a probability. However, many textbooks and DoE software packages encourage users to optimize multiple response processes using the overlapping mean response ("sweet spot") approach. Unfortunately, such a region will harbor factor level combinations that have a low (usually much less than 50%) chance of meeting all of the quality specifications. As such, some authors (including this one) have argued that it is much better to use a predictive distribution approach to optimize multiple response processes. This roundtable will discuss statistical and computational issues regarding this important problem.