There are many reasons to perform design of experiments (DOE). From a machine builder’s point of view, it can help to sell the machine, and it certainly helps to reduce the risk during the design-and-development stage. Many times the machine builder wants to know if the design will work before spending too much time on the project.
Create a quick concept design, make a prototype, assemble and test it. After that, it seems easy to come to a conclusion or make a recommendation regarding the results of a DOE, but when I asked the engineer a few questions about the test, things usually got really quiet. What was the purpose? Where's the data? How did you collect the data? Can you do it again?
Did you hear that pin drop?
On the other side of the scale, there are those who are incredible creators of data who never really accomplish anything. They did a bunch of design, and they collected information and data, but they never made a decision based on the results.
Also read: Design collaboration tools with more
A well-planned DOE is one you will be able to run again in a year and get the same results. A clear scope, drawings, specifications, logistics, test sequence and, most importantly, data collection are required to reach a consistent and repeatable conclusion.
A well-done DOE needs to answer critical design and risk reduction questions. Will the mechanical design and the control methods work in the production equipment? Many times the concept or design doesn't work the first time, and it needs to be optimized.
A DOE or feasibility study using the scientific method and changing only one thing at a time is probably not the way to go when developing equipment, unless it's a well-funded government project. For other manufacturing industries, start with a good concept and breadboard it up for testing of only the critical processes. Most of the design will work just fine, but identification and boxing in the risk, with well-defined DOE scopes, are a good start.
Investigate the cycle times, vision system accuracy, placement x, y, theta and cut-to-length accuracy, press force repeatability, effects of dirty parts, servo motion profiles, robotic handling, feeding, cutting, placing and effects of heating. The scope is to answer questions: Will it work? What is the goal of the DOE?
A poorly planned DOE or feasibility study often produces excellent results but incorrect conclusions. Been there and done that. A good example of this is a system I did that transferred rolls of medical gauze through tubes between the winding process and packaging process. There was a single diameter but multiple width rolls—2, 3, 4 and 6 in.
We set up a tube with defined bends mimicking the final process and started shooting rolls of gauze through it. The 2-, 3- and 6-in rolls worked great; there were no jams; and the rolls didn't unwind. The expectations matched the results. Pass. Done. We built it—hundreds of feet of six parallel lanes of gauze transfer tubes. The first roll we tested in the production system was the 3-in product. Would you believe that 3-in rolls oscillated in the tube and wouldn't move, sinking the whole process?
This failure highlights the need to carefully reference the part drawings, specifications and product configurations during the DOE. This also helps during factory acceptance test (FAT) and site acceptance test (SAT).
The logistics are also important. The equipment configuration should be well-defined. Using temporary or borrowed hardware is a good cost-savings plan, and many vendors are happy to help. Please show your appreciation by using their hardware in the production unit. The customer also needs to be involved. Even though 3-in rolls were not available for testing, guess whose problem it was?
A clear, detailed test sequence and plan for the engineer or technician to follow is important for repeatable data. If you have poor control of the test here, the DOE is a waste of time. Spend the time to document the steps.
Any time spent defining a DOE is well spent, especially when defining data collection. What is the raw data? How much and how often is it collected? What does the data mean, and how is it recorded in an Excel file? Clearly define the data-recording method and the data itself, and you will eliminate one of the more common problems with DOEs—where's the data, and what does it mean?
Recording a bunch of data and then revisiting it in a month to back up a conclusion or recommendation is very difficult if the data is not clearly defined. Perhaps the data results matched the expectations on the day the test was run, but without proper charts, images and data definition, it may not tell the same story next time, when it is reviewed.
While it probably seems like a lot of trouble to scope out, define the test sequence, collect results data and provide a conclusion or follow-up activities; it is time well-spent in successful, profitable machine and control design projects. Define the data and get collecting.
Homepage image courtesy of cooldesign at FreeDigitalPhotos.net
