Why learn classical DOE?

Two-Level Fractional Factorial Designs - click for printable pageCarol Kavanaugh teaches classical experimental design methods for the following reasons:

  • No assumptions about presence of interactions are necessary before running the experiment (if you’re running an experiment to identify critical factors, why should you have to make assumptions about interactions when you don’t know the magnitude of main effects?)
  • We believe it is the presence of interactions in complex industrial processes that often make prediction of future process behaviour difficult or even impossible – hence, in experimentation, it is vital to be able to clearly and independently investigate the effects of interactions!
  • The sequential nature of classical experimentation allows for the building up of increasingly complex empirical models to predict future process behaviour – from linear models to polynomial and complex nonlinear models based on underlying process understanding (form of mathematical relationship among variables). Empirical models can be displayed – graphically visualizing relationships among variables.
  • Life being what it is, experiments are rarely run exactly as they are designed! So, we believe the experimenter should understand how and why each run is included in a design – so if an when the experiment can’t be run as the design says, maximum information can be obtained. Classical methods teach the experimenter how each design is generated, and the inherent tradeoffs in selecting a design. You don’t get something for nothing!
  • Because classical experimenters understand the tradeoffs in running an experiment – there is no need to consult complex Interaction Tables and Linear Graphs.
  • Sometimes Taguchi designs can lead to experiments with large numbers of runs – for example investigating the effects of 5 factors with 3 noise factors, would require:
    • Using Taguchi design: L8 within L8 = 8 x 8 = 64 runs
    • Using classical Fractional Factorial design: 28-4 = 16 runs; 8 main factors would be estimated clearly; if interactions are found to be significant, then another 16 runs could be run to estimate them independently.
  • Powerful, easy to use software for classical design and analysis of experiments is readily available – allowing for such things as missed runs, simultaneous optimization of several responses, design and analysis of mixture experiments, and much more!

Classical experimental design and analysis is easy and fun to learn! – learn how to set up, run, analyse and present designed experiments, at our Design of Experiments Workshop! See what others have said about our workshops!
Compare Taguchi vs Classical DOE, and see Taguchi and Classical Designs (they’re the same!).

Learn how to set up, run, analyse and present designed experiments, at our intensive hands-on Design of Experiments Workshop.