Adaptive DOE Technique

An optional input to the Adaptive DOE technique is a DOE data file (DOE_Results.txt) containing points from a previous run. The Adaptive DOE technique avoids the design points in the data file while generating the next set of data points. You can create the data file using any method; for example, using a previous iteration of the DOE, using an Adaptive DOE optimization or a Monte Carlo adapter, or from a set of user-defined points. The format of the file is a tab-separated history file. The first line is a tab-separated list of parameter names, starting with Run Number, and subsequent lines are tab-separated lists of values.

Executing a series of space-filling DOEs with a projection of the new DOE center and levels is an effective optimization strategy. You can select "clusters" of factors that are modified in parallel. You should place a set of coupled factors in a single cluster. By default, all factors are placed in a single cluster. A run of the Adaptive DOE is centered around a projection of the improvement from the previous DOE. The DOE search space shrinks and expands from run to run to produce a more accurate optimum.

The Adaptive DOE technique is well-suited for the following use cases:

• When you have discontinuous or discrete design spaces and for long running simulations.
• When you know the coupling of factors by prior experience.
• When you can leverage the post processing of the results to control the next DOE center and range.
• When you want to turn parameters on and off for increased search efficiency in subsequent iterations.

Other DOE techniques are more efficient if you already know the behavior model form; for example, quadratic.

The following figure shows how the Adaptive DOE technique distributes points more densely in the interesting area of the design space: