Experimental Designs


Thread Starter


I need to run a trial to find the best conditions for the best paper formation. We know the primary parameters that affect formation. The parameters may vary in significance for different grades. What type of design would be appropriate? Should "grade" be treated as a categorical factor? Please point me to a good reference. I've done designs in school but not in industry. Terry Dixon IP
Your project sounds like a good application for Evolutionary Operation (EVOP). In its simplest form, one gathers data at different operating points around the normal operating conditions. Then one moves away from the worst measured result. This is ideal for something like formation which is often measured subjectively, and most paper industry people can usually tell you the worst from a set of samples, not the best. The British Paper and Board Industry Research Association (now called PIRA) did some work in this area in the 1960s, but I could not find anything on their Website. I have a copy of an article that was published in 1966 that I could scan and send you if you wish. The Swedish company Multisimplex KB http://www.multisimplex.com , have a very advanced computer package that takes this technology to a very advanced level Peter Green Semi-retired control systems engineer

John Beaudry

I have conducted many experimental designs in the food industry. I used to use the classic book by Box on Expermental design. Minitab makes it easy if you want a simple program (relatively speaking). Other programs I have seen are much more powerful for visual presentation of data. Of course you know the process: Screening Design - find the knobs Factorial (or fractional factoral) - find the strength of the knobs RSM - Optimize What no one ever told me is that you don't need to learn everything in one experiment. Yes it is a pain to get line time, but it will fail if you try to do to much. Prototype, prototype, prototype. But don't fall into the one variable at a time approach. No one ever told me that it might be difficult to impossible to randomize. Find out how long it will take you to change from one level to another. It might take days, while you are allotted line time for hours. I did this on my first plant RSM, which resulted in nothing randomized, travel and plant time with little to show for it. Do some contingency planning up front. How do you know that you selected wide enough levels, but not too wide. Is there a way you can determine this and adjust your test. The boss, operators, and other engineers have their pet theories. At least throw the boss or an operator theory into the test. Experience often provides them with an intuitive sense of what is important. Keep the remainder of the test, but add it as another condition. Don't believe them, because often you will be dispelling the myths of the plant. Good luck. It is amazing how powerful are the techniques you learned in school. John