Help to determine the most appropriate Design of Experiment (DoE) approach in SLS print

Question

My goal with the study is to examine the effects of three factors on the yellowness of printed dogbone samples in SLS print (please refer to the attached picture for better understanding):

* Laser power, which can be adjusted between each dogbone
* Height of the dogbone in the build chamber (potentially 7 different heights per build job)
* X-coordinate of the dogbone within the chamber (potentially 14 positions per height)

While I can change all three factors, I am concerned about the potential impact of temperature distribution in the build chamber caused by the placement of the dogbones. For this reason, I’ve considered keeping the placements constant to avoid random effects from temperature fluctuations. I have already made some prints with constant laser power in the chamber and with 14 dogbones at different x-coordinate positions in 7 different heights (total of 98 dogbones for each print).

The primary objective is to evaluate the significance and interaction of all three factors (laser power, height, and x-coordinate) and ultimately develop a regression model to predict yellowness based on these parameters.

I would like to minimize the number of build jobs while still obtaining reliable, robust data on all factors and their interactions (a maximum of 6 prints would be prefered).

I would love if you could help me and share your insights on whether a split-plot design would be the most suitable choice for this setup, or if another DoE method, such as Response Surface Methodology (RSM), would be more appropriate? I am also open to any other suggestions you might have that could optimize the design for my specific constraints.

I hope someone can help!

print job

AUStudent · Answer

XX = spectrophotometer, sorry

AUStudent · Answer

Hi Statman,

Thanks for your reply. To answer you questions:
1) I am well aware of the different factors that affect the "yellowness" which is a response that I want to minimize (it is bad to have yellow outcomes)

2) If the front process is the process that happens when the chamber heats up before printing, I cannot adjust this and make it "better"
3) Yes I have studied the measurement process
4) Yes I know about the exact precision, consistency, discrimination, etc. of the measurement system 
5) I have made a pre-liminary study on the factors before the real experiment yes
6) For your 1) - 6) questions: I'm using a XX, dogbone is approx 200 mm in length, 30 mm in height and 5 mm in thickness, distance between each dogbone i approx 10 mm, distance between each layer is approx 40 mm
7) There will be small change between each run yes, also because the raw material can change from print to print since det mix of reused and virgin powder can differ a bit
8) I'm also going to measure mech. prop. and other physical properties, but lets just focus on yellowness for this example
9) I want to know how the height and placement in x-direction affect the "yellowness" - not that the conclusion would be that only one height and one x-coordinate works
10) If I change the laser power between each dogbone and change the placements of all the dogbones between each print (same as build job) the temperature distribution in the chamber will be too complicated to describe and therefore I will not be able to explain with factors affect the "yellowness". That's why the placements are constant, but the temperature is still varied through the change of laser power between each print
11)  Build job and print is the same thing
12) There will be factors as you mention that I cannot control such as material and exact cooling rate

Hope this helps!

statman · Answer

Welcome to the community.  First, there is no way to decide the "most appropriate" experiment á priori.  What you can do is design multiple plans and evaluate the pros and cons of each plan (e.g., potential for knowledge gain vs. resources).

"The best design you'll ever design is the design you design after you run it."

I am not an SME for laser sintering, so my advice may be suspect, but I have lots of questions and comments about your query.

First, your objective: "My goal with the study is to examine the effects of three factors on the yellowness of printed dogbone samples in SLS print".

Is this really what  you want, or do you want to understand what factors affect the SLS printing process?  Is "yellowness" the only response variable?  How would you know the front process is "good"?  What measures would you take?  I think yellowness is an indication of "bad"?  Have you studied the  measurement process? Do you know the precision, consistency, discrimination, etc. of the measurement system?

Have you performed any directed sampling or CoV studies?  I would think this might be useful BEFORE experimentation.  I would want to know the the relative sizes of the following components:

1. Measurement system

2. Within "dogbone"

3. Between dogbone, within layer (I think this may be the x-coordinate above?)

4. Between layer (I think this may be height above?)

5. Print  run-to-print run

6. Lot-to-lot of the raw material

Knowing this would be extremely help fun in identifying what factors to include in the study, over what noise the study should be run, what are the appropriate response variables and what is the capability of the measurement system.

I don't understand your 3 factors.  I understand laser power and perhaps the speed of the laser moving over the material, but height and x-coordinate I don't get.  You don't want to complete the experiment and find out x-coordinate matters and the process only works for one section your machine do you?  Also you don't want to find out height matters and only one height works.  This will greatly impact the number of dogbanes made per run in production.  I do believe there are likely other factors (e.g., cooling rate) that you could study.

For experimentation, realize this is an iterative process.  The likelihood that you have selected all of the critical factors and tested them at the optimum levels in the first design is near zero.  So, your first design will be done to design a better experiment.

"I am concerned about the potential impact of temperature distribution in the build chamber caused by the placement of the dogbones." This could be studied before the experiment or you can pick systematic locations within the chamber and measure the Y's in this multiple locations. "For this reason, I’ve considered keeping the placements constant to avoid random effects from temperature fluctuation"  This is seldom a good idea.  This takes the effect of the factor (temperature) out of the study.  It narrows the inference space.  That is, your conclusions will be limited to the placement used in the experiment.

I don't completely understand this: "I would like to minimize the number of build jobs while still obtaining reliable, robust data on all factors and their interactions (a maximum of 6 prints would be prefered)."  What is a build job and what is a print?   3 factors and all of their interactions requires a factorial design 2^3, 8 treatments minimum.  Perhaps you don't need that much resolution to start?  You could fractionate, but that would result in res. III.  Maybe you have a model in mind that could minimize the treatments (e.g., don't care about the 3rd order interaction).  Custom design could help to do this.

My bigger concern would be about noise.  Noise being the factors you are not willing to control either because you don't have a means to control them, they are too costly to control or they would be inconvenient to control.  For example, Material (which I assume is a powder) size, shape of particles & moisture in the material. Another example is the ambient conditions (temp and humidity).  You should have some noise strategy (e.g., blocks, random replicates, repeats, etc.).