We use DOE for reducing the effects of uncontrollable factors. What happens in cases where conducting the experiments is much expensive. I know we can minimize the number of experiments, but in cases of semiconductor manufacturing, the cost of conducting even 2-3 experiments is like risking millions of dollars. What alternate options do we have? Does anyone have a real-time scenario?
There is a branch of DOE, Computer Experiments, which include huge literature talking about this since 1979. You may refer to “Computer Experiment” in wiki to get some ideas and references. The basic idea is first to fit a metamodel on few data points. Then, you experiments can be conducted on the metamodel instead of true experiments.
In ISYE 575, we talked about using a fractional factorial design when a full factorial design of experiments are too costly! The main idea is that the design evaluates only a subset of the possible permutations of factors and levels. This method would definitely be something to consider for it makes the experiments cheaper and faster to run.