The Great 3D Printer Experiment: Results of the Second Test

Science is a tedious, messy business. Even doing a simple, rudimentary study like “Great 3D printer Experiment” is surprisingly difficult. Not because my propensity for causal thinking was exposed for the wart-ridden monster that is, or because of the resources and time commitment involved, but because it is nearly impossible to do right. From the onset of the experiment, it was evident that not enough controls were built in (is PLA moisture absorption a factor? I think so) and some selected variables simply didn’t have an impact on the outcomes (something that I should have foreseen). Don’t get me too wrong here; this may not be a “by the book” study, but it did improve my understanding of 3D printing and design of experiments. We all have to start somewhere, so, enough blathering on my end. Let’s take a look at the Taguchi results.

As you might recall from the prelude to this experiment, we were going to run some expanded tests using a Taguchi recommended orthogonal array, which I conveniently reposted below:

Part Two: Taguchi Array
Independent Variables
Run # Layer Height Print Speed Print Temp Outer Shell speed Quality Value
P1 (base) 0.2 50 215 30
P2 0.2 70 210 10
P3 0.2 80 200 20
P4 0.3 50 210 20
P5 0.3 70 200 30
P6 0.3 80 215 10
P7 0.1 50 200 10
P8 0.1 70 215 20
P9 0.1 80 210 30

I suppose I need to say something about Taguchi methodology. There is actually a perpetual hidden war being waged among opposing factions of engineers. On one hand, there are the followers of R.A. Fisher and his “Design of Experiments” methodology. On the other, there are those that embrace the prophecy of Genichi Taguchi and his improvement on Design of Experiments.  For the most part, this war is being waged in the quiet confines of laboratories and computer enclaves, where pocket protectors and protractors are the weapons of choice, all of which are hidden from the view of the mainstream populace. Every once in a while, evidence of the war surfaces. Just look to uncited statements that pop up (then disappear) in the DOE and Taguchi Wikipedia articles. I’m only jesting of course; I ended up using an orthogonal array for this experiment after it was recommended by my friend Andy and then reinforced by the notion that if I were to test the variance of every factor and level presented in the scope of this experiment, I would have to execute 81 runs (3^4). I don’t have the time required for such a thing, nor the amount of filament, so the orthogonal array allows me to compress the tests runs down to 9.

test prints.jpg

Ordinal ranking of our test prints based on time of print.

Like the first set of runs, I gathered the boy, a magic marker, the prints, and some index cards. We used the same simple system for determining the quality of a print that was used during the first set of runs: Each sample was evaluated based on 3 unweighted subjective criteria: sharpness of corner, number of anomalies, and smoothness of face. Each criteria was assigned a number from 1-5; 1 being the best quality and 5 being the poorest. The sum was taken of all three combined criteria to determine an overall score (3-15). The best possible score is 3 while the worst possible score is 15.  Results are below

testresults.jpg

Ordinal ranking of test prints based on print quality.

Interestingly, we had more ties among the test runs than we did during did the previous set, but the king still seems to be layer height. The inconsistency of the P9 print is more than likely due to the quickness of the outershell print; speed tends to create anomalies.

Part Two: Taguchi Array
Independent Variables
Run # Layer Height Print Speed Print Temp Outer Shell speed Quality Value
P1 (base) 0.2 50 215 30  8
P2 0.2 70 210 10  8
P3 0.2 80 200 20  7
P4 0.3 50 210 20  8
P5 0.3 70 200 30  8
P6 0.3 80 215 10  9
P7 0.1 50 200 10  6
P8 0.1 70 215 20  6
P9 0.1 80 210 30  7

 

p9.jpg

The P9 print may look to be the best, but the anomalies of a “fast” print brought its score down.

p8.jpg

P8 is an example of a quality print.

Alright. So we have two tests completed, one to go. I am holding off on drawing conclusions until I can pull the data for the final test. For the next test I am going to run the orthogonal array tests with a slightly more complex model. Behold, the 3D hubs Marvin the Martian. This model, recommended by the thingiverse forum, is a more difficult print than cuboid. Using this model should stress the printer while also confirming or denying the results from the first two prints.

marvin.jpg

I can’t wait to see what happens. To be continued in the final installment of “The Great 3D Printer Experiment”.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s