Expert system for statistical design of experiments

DETAILED DESCRIPTION In accordance with the present invention an expert system is provided which can interface with a client to generate and analyze balanced designs using a linear models approach, expert reasoning abilities, and existing software packages.
The expert system of the present invention emulates the role of a consulting statistician, interacting with the client to gather information needed to generate the descriptive mathematical model and physical layout of the experiment.
The role of the expert system of the present invention is to bridge the gap between the user, an engineer who is inexperienced in statistical techniques, and the SAS package.
It interacts with the user to identify the goal of the experiment, the input factors involved, how they relate to one another, whether there are nested factors (i.
e.
factors whose levels depend on the levels of other factors).
Once the necessary information has been collected from the user the expert system writes appropriate SAS code to generate an experiment design.
The results returned from these procedures will constitute enough information to produce an experimental design.
However, before producing a data collection sheet, the expert system evaluates and refines the design through the use of a mathematical model.
The expert system advises the user about any problems with the design, presents alternatives for improving the design, and identifies advantages and disadvantages of each alternative.
Once the user is satisfied with the tradeoffs associated with a particular experiment design, the expert system generates the data collection sheet of that design.
When this data collection sheet has been filled in with actual experimental results, it requires only a simple call to a few SAS procedures to determine which particular input factors or interactions affect the response variable(s).