Process Optimization

Manufacturing processes can easily be optimized . . . Now that is a bold statement which I can back up. Lets assume you need to maximize the performance of your product by such things as:

?Minimizing material cost or

?Minimizing process time or

?Maximizing product appearance to minimize rejects or

?Maximizing product taste

Notice that the claim is not just reducing or improving. That implies 'just' an improvement but not necessarily optimized to the maximum. Maximizing and minimizing results are easily achieve provided that the product is manufactured by machines such as plastic molding machines, die presses, and processes using a recipe such as baking bread. Essentially any process that can be repeated and has a minimum of human involvement can be driven to peak performance of the process.

The method is extremely simple to use as a computer program guides you:

1.The computer asks a few simple questions about the process. The questions asked are that the baker, technician, or setup operator can answer. This is not rocket science stuff. The mathematics of it certainly are but thats the computers job.

2.The computer details eight experiments to run. Each experiment involves changing different specific setting of the process.

3.The success of each experiment is judged on a scale of 1(failure) to 10(excellent) and reported to the computer.

4.The computer then gives the best setting for each of the controls.

You must admit that this is simplicity itself.

For the non-technical person, the point is to insist the technical group investigate the process known as Design of Experiments. There are oceans of information on the net about Design of Experiments. Someone in your organization will have to take a short course to obtain the software and how to use it. It is not difficult to learn. I took a two day course and never looked back.

What are the questions to be answered?

?List all the controls the process has. For a molding machine it would be various pressures, temperatures, speeds, and timings.

?For each control estimate the highest and lowest value you think may work. You do not have to be accurate. Just what you estimate the highest and lowest are. It is best that the process actually works, even if not well. I once had a process we could not run at all. We guessed all the values and had wonderful success.

Given this information the computer details eight specific production runs using its specific settings. There is only one proviso: The tests must be run in the order specified for rearranging the order of tests offers a high risk of failure.

Make enough samples for the process to settle down. How good are the parts? Evaluate the results. Place a rating from one(failure) to ten(excellent) for each experiment and report them to the computer.

The computer responds with the best setting possible for the process, as the machine and controls currently exists. Remember, additional controls reduce the uncontrolled variables reducing the range of product variations. Lacking key controls means a grater variability of the production results.

As a bonus feature, the computer reports what settings are critical, which ones are critically interrelated, and which are not particularly critical. Further refinements for other features can also be done.

Probably the most famous story about Design of Experiments was about a Japanese ceramic manufacture that had a wonderful product but was running a ten percent scrap rate. What to do? Purchase a better oven? Would that work? So he ran a Design of Experiment and reduced rejects to one percent with the existing oven. One surprising result: For hundreds of years it was known that one specific chemical was a must have in the formula to make excellent ceramics but it turned out it was not necessary.

So that is the basics. I have used it with wonderful success.