Without DOE, you're stuck with the world's slowest method for success-trial and error.

With Design of Experiments, you just have to test at the high (+) and low (-) values for any particular "design factor" (e.g., pressure, temperature, time, etc.) from your QFD House of Quality, not every increment in between. And you can test more than one factor at a time.

For more info, see: http://www.qimacros.com/lean-six-sigma-articles/design-of-experiments/

Download the 30-day trial at: http://www.qimacros.com/trial/30-day/

This video shows you how to perform DOE analysis in Excel using the QI Macros.



Design of experiments can help you find the optimal settings for a machine or temperature or whatever without a whole lot of trial and error.

You can imagine I take hundreds of trials to try and figure out the right combination of ingredients to give you the optimum setting to produce an injected molded part or even to come up with a way to measure the optimal weight as another direct mail piece to formulate that direct mail piece.

So, in the QI Macros, there are a series of templates you can use to help you do a DOE study.

Up here, you'll see there's a number of factors. So, you can insert your factor names here. You can do up to four factors. That's why they call it a two factor, a three factor or a four factor experiment.

You want to set up the high levels and the low levels so you can actually see what's going on.

I brought in some data here that we can use to do this. Let's just take that and copy that in there. So in this case, we have a die, and it's going to have a certain temperature. The low-level temperature is going to be room temperature, and the high temperature is going to be 200 degrees. Pour time—we want to vary that between 6 at the low end—and we'll also try it at 12 seconds.

Now, what you want to do then is go out and conduct your experiments once you know what your high and low values are for each one of these.

Down in here, you'll see how you want to set up your trials. So, trial number 1—and you're going to randomize how you do these—it's going to have a low value, a minus and a minus. So, we wanted in that case to look at die temperature at room temperature and a pour time of 6 seconds. We want to try that.

The next one, die temperature would be high, pour time would be low. Here, it would be 200 degrees and 6 seconds, and last but not least, 200 degrees and 12 seconds for pour time.

So, these are the key things you want to look at in terms of setting up each test, high and low. What I've done is I've collected some data out here that we can use—let's copy that in. Now, we have some data from our studies.

We want to go down and look at the charts to see what's going on here. So, you can see there's from low to high temperature, room temperature to 200 degrees and pour times 6 to whatever and whatever this metric is.

Then, we come down and look at the interactions between these two, and here you'll see that there's a significant change in this process. Our optimal point for this will be right at that intersection. That looks like it might take instead of 6 or 12, it might take about 10 seconds—that would be the optimal pour time. We want our die temperature to be high.

So, this is a way to quickly be able to hone in on exactly the way to set up your system so that it will work easily.

I also brought in an example of one for DOE injection molding, and in this case, we had seven different factors using the L8 to Guchi Matrix. Again, you'll see all the different setups for each one of the trials—same thing with putting in your data.

Up here will show you where the interactions are, well one and three injection speed and melt temperature have an interaction. One and five and one and six injection speed and holding time and cooling time. It appears that one thing interacts with a lot of different things.

Go down here, you can see where the different piece parts of each one of these are. We come down and look at how the factors interact when they're in parallel, there is no interaction. So, factor number one and factor number two really have no interaction; however, one and three, as we said before, there's an interaction; two and three, it appears that at the very low end there might be some interaction, but that's probably the optimal point is low and whatever the temperature is would be optimal.

You can start to see that it's easy to analyze some fairly complex trials of things to come up with the optimal way to select how to run any given machine or any given tests of two different things and come up with the right answers, using the QI Macros and the DOE template.