PRISMTC :: smarter statistical solutions - PRISMTC Tipsheets - Augmenting for Optimization: 1 Introduction

Augmenting for Optimization: 1 Introduction

Design Expert offers you the following choices for adding or augmenting new runs to an existing design:

1. Generate a Foldover, a Semifold-over, or a Factorial D-optimal of your initial design to de-alias 2-factor interactions (2FIs) and resolve any ambiguities.

2. Improve the predictive power of your model for optimization by either adding axial (star) points to transform a screening design to a Central composite response surface method (RSM) design, or using the RSM D-optimal option to select the additional runs needed to update your model.

Augmenting to a Central Composite Design (CCD) is the standard approach. Alternatively, the response surface method (RSM) D-optimal augmentation option can be used to create an additional set of experiments for fitting higher order models. This latter augmentation also allows you to remove unlikely terms and reduce the required number of runs; repair faulty designs as well as accommodate previously performed experiments. A more versatile augmentation tool, but you must handle it with care.

To handle the augmenting of designs for optimization within the Design Expert DX7 software tool, we have further tipsheets to help you:

Augmenting for Optimization: Building the Design
Augmenting for Optimization: Analysing your Results

These tipsheets are linked to this case study

Move
Close

Heat Sealing Process

This case study is based on a Heat Sealer example from Wayne A. Taylor’s book “Optimization & Variation Reduction in Quality”. 1991 McGraw-Hill New York

A heat sealing process forms a secure seal for the blister packaging of a product. The seal must be strong enough to keep the product safe during handling but easy to open by the customer when required. The target strength of the seal is 26 lbs ± 6lbs. Baseline results from 20 batches of 5 samples of packages from a production run suggest that the heat sealing process is in control but not capable due to large variation (i.e., it is consistently producing packages that have an unreasonable likelihood of failing!)

It was decided to acquire Process Understanding before establishing Control and Capability, by considering the effects of eight process parameters not only on the average seal strength of 5 packages but also on the standard deviation. A 28-4 resolution IV (Yellow) screening design identified 4 key factors affecting the dual responses of average and variation in seal strength of a blister packaging process. There was also evidence of curvature for both responses; indicating that we are experimenting in a region over an optimum. Rather than run a fresh set of experiments in this region, you should contemplate augmenting your existing set of data to form a response surface design. This will enable you to carry out an optimization and improve the predictive power of your model in a more cost effective way

return to contents

next >

Posted on Jul 21, 2008 at 05:50PM by

prismtc in Augmenting Designs, DX7, Design Expert, Designs, Optimization |

View Printer Friendly Version