DOEpack release supports multiple analysis, mixture experiments

New features in the forthcoming release of DOEpack, a program that supports experimental design and underpins Six Sigma efforts, make life easier for users, including those who develop mixture designs or need multiple analyses per study.

Jeff Aughton, a member of the PQ Systems development team in its England office, has led the effort to broaden the program’s usefulness and enhance its ease of use. PQ Systems consultant Gordon P. Constable, Ph.D., is the U.S. participant in the program’s improvement. The latest version is expected to be available in early spring, 2005.

Earlier versions of DOEpack allowed one ‘type’ of analysis per study – one ANOM, one Effects plot etc. DOEpack 3.1 imposes no limit on the number of each type of analysis. This makes it simple to compare different settings and different treatments directly, thus leading to a greater understanding of the process. Screenshots from the sample file DCMp234 illustrate this feature:

This is the problem tree for the file. Note that there are several effects plots and also several regression studies – 1 for each type of nozzle used in the experiments.

Using these multiple analyses, one can create the multicharts below that demonstrate the nature of the interactions between the factors and the differences between the nozzles.

A second feature in the forthcoming release addresses mixture designs. The study of mixtures creates special problems for DOE, since the factors are constrained in that they must sum to 1 (or 100%). DOEpack now includes popular mixture designs such as Simplex and Simplex-Centroid. New 2 and 3-factor contour plots, such as the one shown, now complement these designs:

Many real-world applications feature several response variables, and there are often different requirements for these. For example, one always wants to increase quality and reduce cost, but it is rare that the settings that achieve the first also achieve the second – in fact the opposite is often the case. DOEpack includes two new tools to help with this situation. These are the scatterplot (a simple visual aid) and the use of desirability functions (a powerful technique to optimize several responses simultaneously).

Here is a scatterplot:

Each response variable can be assigned a ‘desirability function’ that specifies how important its target and specifications are.

In this diagram, the target is 57.5 and the specifications are 55 and 60. Ideally, the value is near to 57.5, but not less. However, if the value is greater than 57.5, it does not matter if it is near to 60, provided it does not exceed that number.

Once functions have been designed for each response, DOEpack uses a special technique to optimize a composite function to determine the factor settings that give the best compromise solution for the user.

In addition to these three improved features, others to watch for in the new release are new methods of handling analyses, enhanced filters, desirability analysis, and new problem node. Since the last full release, customers with maintenance plans have benefited from the addition of language support, printing options for regression, improved randomized replications, and alternative data entry methods. A new algorithm to solve the regression equation was added in DOEpack 3.0.21.

Release dates for the new program will be announced soon via the PQ Systems website, which also provides information about DOEpack’s features: http://www.pqsystems.com/products/sixsigma/DOEpack/DOEpack.php