As mentioned in my previous post, I had planned on doing some more testing and analysis of specimens from each printer. I’ve now collected some more data, but before I present it all, I want to share one of the findings I obtained regarding the nylon on the Markforged printer. When I first started printing my samples, the Mark One was loaded with a fresh roll of nylon, I got the graph below, which I talked about in my last post.
When I decided to test 3 more samples on each axis, I printed a new set of nylon specimens nearly a month later, and was really surprised when I saw that the stress strain curve had a much different behavior:
As you can see from the plots, the second set of samples fails at a higher strain than those of the first set, although their maximum stress is roughly equivalent. I was taken aback by this change: nothing in my process or setup had changed (leveling the build plate, organizing the parts on the print bed, etc.), and that’s when I realized that the problem lay in the filament. The nylon spool on the Mark One is kept in a watertight, sealed Pelican Case, but over the course of a few weeks and a few people opening the case, it seems likely that some moisture got in. Nylon is very susceptible to moisture and swells upon absorption, and water absorption usually makes materials more extensible, so I came to the conclusion that the nylon had absorbed water vapor and its material properties had been affected.
The table below contains the actual values obtained, with the first set of samples tested highlighted in red and the second set highlighted in green. You’ll notice that the yield strength and moduli are different than the values I listed in my previous post. To calculate those values, I had approximated the modulus and yield strength with a straight edge, but I wanted a more controlled method than my original approximation. Since then, I have written a MATLAB function to perform those calculations for me by approximating the slope on the initial linear region of the curve to obtain the modulus. The yield strength was calculated by creating a line of that slope with a determined strain offset from the origin and finding the strain at intersection of that line with the stress-strain curve.
I wanted to find out whether or not the two sets of samples were statistically different, i.e. the probability that these samples come from identical or different groups. This depends on how far apart the means are and how big the standard deviations of each group are, and can be evaluated using a t-test. The t-test returns the probability that the null hypothesis of the two sets being from the same group is true. Should the probability it returns be less than a stated significance level, then there is a very small probability that the two sample sets are of the same group, and we can consider the null hypothesis of to be rejected, i.e., the sets are not from the same group. We use a significance level of 5%, which means that if the value it returns is less than .05, the samples from my first and my second rounds of testing can be considered statistically different. I ran this test on the values obtained for each axis and across each material property, and here were my findings:
From the test results, it appears that along the X and Y axes the two sets are statistically different. Just from looking at the graphs, you can see that the new samples stretched much more in the X and Y directions than the old samples, which drastically affected the strain at fracture and the toughness. However, all of the values obtained from samples printed in the Z direction are statistically similar. The readings I got from both sets were highly variable, with some standard deviations exceeding 100%. Either this means that the absorbed water vapor in the nylon does not affect the adhesion properties of the nylon to itself (because the material properties of specimens printed along the Z axis rely on layer by layer adhesion), or more likely there was so much variation in the behavior of samples along the Z axis that the material differences were insignificant compared to the variability in data. As I mentioned in my previous post, it is tricky to print the test specimens on the Z axis because they are thin and tall, and often fall over.
From the t-test, it seems that if nylon filament is used, the quality and age of the material plays a big part in the material properties this printer will produce along the two axes that lie along the build plate. The X and Y axes produce parts that are much stronger and more consistent than those printed on the Z axis of the printer, yet they are more susceptible to changes in the quality of the filament. I did not see anything near these trends on any other printers I tested samples from, because the other materials used are not as susceptible to humidity as the nylon filament is. Even on the Mark One with embedded fiber, the fiber adds a layer of consistency to its prints (as I discussed briefly in my last post), so the two sets of Kevlar-embedded samples did not seem to have any statistical difference.
An interesting test to run would be to see if the samples degrade over time when left out as well. If you printed something with new nylon filament, and left it out for a month, would it have similar properties to a part printed new, but with month-old filament? I would expect changes in the material properties regardless, but I don’t know how they would compare to the data I discussed in this post. I’d be interesting to look into!
Although the two sets of test samples did end up being pretty different, when I do my final evaluations across all printers I will consider them all part of the same set because they represent possible variations in what can come off the Mark One. But lesson learned, if you’re printing in pure nylon, and you want consistent material properties, load up a fresh roll and make sure it’s sealed tight!