I frequently perform 2k factorial design of experiments and fit the data to a linear regression to control/optimize processes in a manufacturing environment. I have noticed recently that a large portion of my experiments must be revised fairly early due to faults in my experiment model (e.g. my experiment for dry film thickness was yielding values of 3-8mils and my target value was 1.6mils). This is worrisome because I occasionally want others to collect the data.
Is there a rule of thumb for when or how much exploratory data is needed for this type of DOE? Or any advice on the matter?