A Design of Experiment (DOE) is a technique that helps to determine the influence of the process factors to the process output. It originated in the US after World War I, was taken to Japan in the 1950s by W Edward Deming and became very popular in the late 90’s thanks to six sigmas process improvement methodology.
When analyzing a process, engineers often get overwhelmed by the number of factors that have the potential to influence the response, such as the material properties, the machine settings, the operator skills, the environmental conditions etc..
A common approach is for engineers to perform a few random experiments without a specific methodology, also called trial-and-error, in an effort to modelize their process. The results are often scattered, not easily understood and rarely accurate.
The Design of Experiment methodology provides a way to plan the experiments and aim at a clear objective. It will reduce the number of experiments and above all facilitate a quick interpretation of the results.
At Central Midori we have been using DOEs to determine what are the most important factors of a specific process. The goal is to better understand and optimize our processes to reduce variations and deliver higher quality to our customers.
As an example: our most important process, screenprinting, is empirical by nature (i.e. the result was obtained from experimental observations) and we have applied DOEs to help our engineers better control what otherwise was “tribal knowledge”.
What is the screenprinting?
Screen Printing is the process where ink is transferred through a mesh screen onto a surface. The stencil blocks the ink from passing through the mesh in certain areas and where it passes through it makes the printed image. The squeegee travels across the stencil, the screen makes contact with the substrate and ink is transferred through the openings.
It is a complex interaction of a multitude of factors including the squeegee material, speed, angle, pressure, mesh material, thread diameter, mesh opening, stencil thickness, ink viscosity. The tendency is often to rely on the printer experience and sacrifice process control. That’s where a DOE can come very handy.
The DOE has two types of objectives: 1) to identify, among all the factors influencing the process, which are more contributive and then 2) to determine the levels of these contributing factors to reduce the process variation.
Going back to our screen printing example: until screen printing was being used in electronics the ink thickness was never an important output. It all changed with resistance controlled requirements and became central and paramount to setup and control the process.
A few years ago CMI conducted a DOE to study which factors had the most influence on the ink thickness once deposited onto the substrate. The results showed that the mesh and the stencil attributes were significantly more important than the squeegee pressure. Intuition told operators, on the other hand, that pressure was the key factor. Since then engineers are taking full responsibility for ink thickness by calling on the screen characteristics.
A factorial design will analyze all the possible combinations. To continue the example of the screen printing process, if we study the three following factors:
- Mesh opening
- Thread diameter
- and stencil thickness
and each of these factors have 2 settings
- Mesh opening: narrow and wide
- Thread: thin and thick
- Stencil thickness: low and high
Then there are a total of 8 different ways of combining them (2³).
Now let’s say that we plan to set 3 levels (low, medium, high), then the number of experiments will increase to 27 (3³), becoming a lot more cumbersome.
The question is now: is it truly necessary to perform all these experiments to find the influencing factors?
Scientists have developed partial or fractional factorial design to reduce the number of experiments, one of them is the Taguchi experimental design. If you are interested in learning more about Taguchi I recommend The Design and Analysis of Experiments by Douglas C. Montgomery.
Things can get even more complex with the introduction of the combined effects, also called interactions. Assume that we study three factors (A, B and C) with two levels (1 and 2).
In a phenomenon without interactions we notice that when factor A changes from level 1 to level 2 the effect on the response is independent of the levels of factor B and C. However in the case of interactions between factor A and factor B, the effect of factor A on the response is not the same depending on whether factor B is on level 1 or level 2.
Interactions are generally added to the design of experiment as additional factors.
Design of Experiments are not easy and I apologize to our quality manager for trivializing some very complex techniques.
Today a majority of our engineers are capable of conducting simple DOEs without the assistance of the quality department and besides the capability analysis, the Design of Experiments has been of great help to understand our processes.