Design Of Experiments (DOE) with JMP®

When your goal is to create an experimental design that takes into account specific parameters like time, budget and other considerations, the unique Custom Designer in JMP constructs a design customized to your problem (using an optimal design), so you don’t have to force your problem to fit a prescribed textbook design.

The Custom Designer always makes the best use of your experimental budget. Using its computer-generated designs allows you to tackle a wide range of challenges, all within a unified framework. You can include continuous, multilevel categorical and mixture factors within the same design, and specify hard- and very hard-to-change factors for automatic creation of the appropriate split-plot, split-split and strip-strip designs.

Also, the capability to define factor constraints, model effects and interactions, as well as include center points and/or replicate runs as you build your design. Finally, the Custom Designer allows you to perform sample size and power calculations, as well as visualize alias structures all to aid you in determining whether your experimental investment is likely to be worthwhile through rich design diagnostic capabilities.

The Custom Designer allows you to builder smarter designs more quicker and efficiently to save you time, effort and make better use of your resources for conducting experiments.

Ronald Fisher first introduced four enduring principles of DOE: The factorial principle, randomization, replication and blocking. But until relatively recently, generating (and then analyzing) a design to exploit these principles relied primarily on hand calculation. Despite this burden, the ingenuity of practitioners over more than 80 years has led to a series of widely applied design families adapted to meet specific situations and experimental objectives. JMP offers all of the classical design types you would expect, including Full Factorial, Screening, Response Surface, Mixture and Taguchi Array. After defining factors and responses, JMP lets you pick an appropriate design from those listed and provides various design evaluation tools, such as prediction variance profiles and FDS plots, to assess your selection before committing any resources. Once the runs have been conducted, analysis is straightforward thanks to the pre-built JMP scripts that are stored in your data table during the design process.

Even when there is no intrinsic variability in the response, DOE still finds application in exploring highly dimensional factor spaces efficiently. To meet this situation, JMP provides Space-Filling designs, which are typically analyzed with the Gaussian Process smoother to make a surrogate model with low prediction bias and variance. JMP can also generate and analyze Choice Designs in which consumers or users are asked to state their preferences between alternatives, including price as a factor if desired. Finally, JMP provides designs for Accelerated Life Tests and Nonlinear models. And if needed, you can add more design families to JMP through its scripting language, JSL.

Although vital, design is only half of DOE. No matter which design you decide to use, JMP makes the subsequent analysis as easy as possible. Depending on the situation, the table containing your design will automatically contain the right script to analyze your results, usually via the Screening or Fit Model Platform. With multiple responses, you can simultaneously fit different models with Stepwise refinement using a chosen stopping rule. When you have built models you think are useful, The various Profilers in JMP allow you to interactively work with them and visually identify viable operating regimes and factor set points. No matter how complex your problem, The built-in Optimizer in JMP can perform the inevitable trade-off between responses with a single click. Once you have the sweet spot, you can then use the integrated Simulator to see how robust this is likely to be in practice.

Hard-to-change variables, such as the temperature of an industrial oven or the location of a cornfield, exist in the real world. A completely randomized design might require such factors to be reset with each run. Clearly, this would be impractical or cost-prohibitive. The designed experiment most appropriate for such situations is called a split plot. JMP produces I-optimal split-plot, split-split, and strip-strip designs. JMP generates the design and includes the appropriate random-effect restricted maximum likelihood (REML) model as part of the table that contains the experimental design.