Getting Good Data from DOE

How do you know what process variables control a part’s dimensions or other properties? A properly conducted DOE is the answer—but incorrect procedure will yield useless data.

Related Topics:

The injection molding process encompasses hundreds of variables. There are so many that some practitioners let science fall by the wayside and take the attitude that molding is an art.

Honestly, sometimes I too get a bit frustrated when I cannot explain what is happening. But there is an answer—we just have not found it yet. Often the question is: Which process variable or variables is/are controlling some critical aspect of the part? This aspect could be a dimension, warp, gloss, or whatever.

Dealing with so many variables can get confusing. One way to get a handle on them is to organize or categorize the processing variables into groups. To do this I use Don Paulson’s four plastic variables (Don founded Paulson Training Programs Inc.), but I change the word “variables” to “categories.” This helps separate plastic variables from machine variables. So I review the process problem relative to four categories of variables—from the point of view of the plastic: flow rate (encompassing shear rate and injection velocity); temperature; pressure; cooling rate and time.


To answer the question of which processing variable is influencing part dimensions, warp, gloss, ovality, etc., many like to perform a Design of Experiment (DOE). In a DOE one selects one or more “factors” (process variables like injection velocity or hold pressure), and then assigns levels to test for this factor. Levels are the high and low for a particular variable. For example, choosing the factor of injection velocity, a low level might be 0.50 in./sec, and the high level might be 2 in./sec.
Care has to be taken in choosing factors and levels. Keeping both to reasonable numbers can save huge amounts of time and money. Processing factors or variables that may be important should be identified during Operational Qualification (OQ) for medical molders and Production Part Approval Process (PPAP) for automotive processors. Levels should also be established in this initial work-up of the mold, resin, and process. If we pick injection velocity as a factor, the levels can be found by looking at the viscosity curve and parts at each tested velocity.

Whatever factor you pick, first you have to ascertain if it’s a machine or plastic variable. If the former, you must evaluate it with respect to the four categories listed above to help you organize the plastic variables.

So let’s suppose we’ve chosen injection velocity as a machine variable, and we have a low and high level for velocity from the viscosity curve. Now subject the factor of fill time to the four questions. With injection velocity/fill time we change the flow or shear rate. So what are the ramifications on the plastic? Changes in shear rate will change viscosity and perhaps fill pattern. See the influence? In this case velocity is the machine variable and fill time is the true plastic variable.

Restate the two levels as fill time by finding what the fill times were for the two injection velocities. Fill times can be reproduced from one machine to another, but duplicating velocities will not duplicate plastic conditions. Let’s say the high/low fill times were 3 sec and 0.75 sec. A molder might think all he needs to do is set injection rates of 0.50 in. and 2.0 in/sec in order to achieve those fill times, but the experiment will provide incorrect results.

Why? If the molder changes fill time only by changing the injection velocity, without touching the shot size or position transfer set points, he actually obtains two different shot sizes. If he sets the same stroke transfer position for both levels, watch what happens—again, from the plastic’s point of view. Using the slow velocity first to provide a 3-sec fill time, the molder will find a 0.60 in. transfer position provides a 98% full part by volume. He sets up this transfer to second stage to pack out the part.

Now, he increases the injection velocity to whatever machine setpoint is needed to obtain the 0.75-sec fill time. What happens if he uses the same cutoff position of 0.60 in.? Well, I would not want to be around for this shot if it is done in a tool with a slide in it.

All machines are governed by the laws of physics, and even though our hypothetical state-of-the-art press still transfers at 0.60 in., the ram/screw now has significantly more momentum, kinetic energy, and will coast beyond the 0.60 in. setpoint farther than it does at the slow fill time. This might flash that slide/mold. Even if it doesn’t, the first stage does not make the same 98%-full part, and the same pack and hold pressure will provide significantly different cavity pressures. Yes, you’ll have parts, but the data will not be apples-to-apples for appropriate comparison.

To get good data the processor has to re-establish the position transfer for fast fill to achieve the same 98% full part as with the 3-sec fill time. Note that we have only touched on one of the four questions; do any of the others demand a change in molding procedure? Do changes in fill time influence plastic temperature, pressure, and cooling rate or time? How do you perform the DOE to get good data?


Another common error in setting up DOE is to set levels arbitrarily. For example, suppose a molder wanted to test hold pressure as a factor and picked low and high levels of 400 and 800 psi hydraulic hold pressure, respectively. There are two significant problems with these levels. First, they are stated in machine pressures when they should be stated in plastic pressures. (My bet is that 80% of DOE variables are stated as machine variables instead of plastic variables. Always work with plastic variables.)

Second, can the mold be packed properly with the low level? And would trying the high level damage the mold by flashing a slide? Anybody want to fess up and admit a mold was damaged attempting an inappropriate level? I estimate that more than 50% of the DOEs performed have levels incorrectly chosen. (Ever wonder why DOE data don’t always make sense? This is one possibility.)

For hold pressure, it is an easy and fast experiment to find the low and high limits possible for the resin, mold, and process. After determining gate-seal time, and you have established whether the part needs gate seal or unseal, you set the hold time. Then start with 500 to 1000 psi of hold pressure and slowly raise the hold pressure until parts look like they might pass QC. Then continue to raise the hold pressure, slowly, until you get early evidence of flash, sticking, pin push, stress whitening, or any mold-opening or part-ejection problem.

You cannot run near this high pressure, so back off a few hundred to 1000 psi plastic pressure for your high limit. You now have established the low and high possibilities for this mold, resin, and process. And you haven’t busted a mold because somebody plugged in a number dreamed up in an office.

Editor Pick

“Unexpected” Consequences: Expert Discusses Impact of Lockout Tagout Standard Changes

The decision on July 18, 1996 by a federal appeals judge to vacate three citations against General Motors reverberates anew as interested parties await the fate of one word—unexpected—in a long-established standard.