Our First Test Went HORRIBLE | Testing 3D Printing Filaments
Вставка
- Опубліковано 7 лют 2024
- In our inaugural video from Tangled Testing, we learned about the challenges of gathering reliable data on 3d printed parts, further emphasizing the inconsistencies of 3d printing materials.
Our tests were performed by compressing on cantilevered beams with varying wall thicknesses and infill densities. Surprisingly, increasing infill and wall thickness didn't always equate to stronger parts, indicating material quality issues.
This video not only highlights the critical need for rigorous testing of 3D printing materials but also sets the stage for Tangled Testing's mission to provide reliable data and insights. Through increased sample sizes and the use of higher-quality filaments, we aim to isolate variables such as infill and wall thickness to determine their true impact on material strength, moving beyond the initial challenges posed by inconsistent filament quality.
🔔 Subscribe to Tangled Testing for updates on our latest videos and discoveries in the world of 3D printing. Together, we'll unlock the full potential of 3D printed materials and ensure your projects are built on a foundation of reliable data.
About Tangled Testing
Welcome to the official UA-cam channel of Tangled Testing - your go-to source for testing and analysis of all things 3D printing. Our mission is to shed light on the true quality of 3D printing materials through various testing methods. Whether you're a seasoned 3D printing enthusiast or just starting out, our channel offers valuable insights that will empower you to make informed decisions about the materials you use. Subscribe for insights that will enhance your 3D printing projects and join us on our journey towards higher standards in 3D printing. - Наука та технологія
I recommend trying to follow an ASTM procedure if at all possible...perhaps D790 for what you are trying to investigate in this video...because that will help you avoid a variety of non-obvious problems. For example... (1) A circular cross section in flexure will expose the small portions of material at the top and bottom of the cross section to the highest bending stress, so any flaws there will dramatically influence your results. (2) A short cantilever will have both high bending stress and high shear stress, which will prevent you from investigating only one cause-and-effect. (3) There is a nonlinear relationship between the bending stress and the wall thickness, so you "should" see hardly any change in the failure load once the wall gets pretty thick. Think back to the "neutral axis" concept you learned about in mechanics of materials, where the middle of the cross section experiences no bending stress...which also causes the infill to have less of an effect than one might think. Etc. To avoid these engineering-mechanics issues, D790 recommends a rectangular, simply-supported beam that has a span-to-depth ratio of at least 16.
I second this and also recommend you throw the simply supported beams into an analysis package like Ansys to see what the stresses look like throughout cross sections with differing infills and wall thicknesses. It'll be a great way to demonstrate to viewers the input of each parameter on beam strength. Anyone else having bittersweet flashbacks to shear and bending moment diagrams?
I doubt the guy running the channel has a Mechanical Engineering degree. His approach to all of his testing and methodology, from this channel and his other channel all seems very half-arsed and lacks concrete scientific reasoning. I wouldn't trust this channel too much.
You made a big deal early on about using standard test coupons and test equipment. So your first official video is testing a random homebrew design with unknown filament as an indirect way to measure layer adhesion. You say the results seem to defy physics, but I don’t think you understand the physics of bending a cantilevered beam made of anisotropic material. Other commenters have already pointed out things to consider, but it boils down to the fact that you are exposing too many variables in a single test.
Yea this is clearly lacking an engineering mindset. If you compare it with CNC Kitchens tests this really feels like amateur hour, sorry
Agreed. But it does help show how many variables do pop up with 3D printing. And hopefully over time they can come up with guidelines for us all to print with.
I doubt the guy running the channel has a Mechanical Engineering degree. His approach to all of his testing and methodology, from this channel and his other channel all seems very half-arsed and lacks concrete scientific reasoning. I wouldn't trust this channel too much.
@@donguyengiac5046I believe he is a mechanical engineer
I'm going out on a limb and suggesting that the surprising results may not have been material properties changing throughout the reel of inexpensive PLA. Adding wall thickness could have increased internal stress that reduced interlayer bond strength. I look forward to the same test with Tangled or other quality material to see if the surprising results persist. When you're surprised by an experiment, you are learning. I have a feeling that Slant 3D is going to learn a lot from Tangled Testing. We all are. Thank you.
If you're testing layer adhesion then, keep in mind that the longer a layer takes to print the more cooling the previous layer gets, which might affect the layer adhesion.
This is the reason and I think it was CNC kitchen who also concluded that
Yeah this was my very first thought too.
This is why you don't print at marginal temperatures like 210-215 where your adhesion to the previous layer depends on it still being warm or the nozzle moving slowly enough to reheat it well. Crank that to 250+ and you will never have layer adhesion problems. The ONLY reason not to do this was PTFE lined hotends which are long since obsolete.
Yeah go print PLA at like 260...
@@rrrlasse2 That's my daily driver temp for normal printing. The Chubby Croc challenge you can see on my channel was at 310°C. That'll smoke if you leave it idle, but 260 can sit pretty much indefinitely with at most some discoloration after many minutes.
Thicker walls means larger compression/tension force between the inside and outside on a surface.
Which means a higher shear force being exerted along the layer lines at any specific force.
It makes sense to me that a 3d printed part with layer lines would actually see a decrease in strength at specific layer line wall thicknesses increase.
If you're printing a cylinder, how do you align the infill to be the same direction every time? I read a couple other comments about using a rectangle, and within doing a rectangle you could also do a consistency test with iterations where you rotate the part say 5 degrees at a time up to 90 degrees and take an average of the parts to see if even just the orientation of the rectilinear infill relative to the outside shell affects the strength of the part in the dimensions that you are testing, changing no other variables.
Please, make sure to include error bar in the result graph. It's a good way to visualize the variation introduced by both the test methodology and the filament.
Might be important to have the items placed the exact same orientation as the previous tests to limit variability that can arise from the inner infill patterns starting in different orientations. Another thing to look at is maybe a good idea to have a really good filament being tested as a standard. Because were these tests tested with a reputable filament brand? What did those results look like? If you were to say get the same results with a good filament brand. Then the test itself is flawed and not the filament. Also, you need to do significantly more tests than a single instance. You can have variability between parts, even with the same infill, perimeters etc. There is a reason why in the research community we use heavy statistics and have 1000's of subjects/iterations to test single variables and come to a conclusion. There will always be outliers. There will always be ranges.
Orientation is exactly a thing I thought about! Thank you!
@@ThePajadamPL Me too. Just need an arrow at the end to get the same orientation
More walls mean longer layer time, means more cooling per layer, means less bonding. A good way to overcome that would be to anneal all parts before testing. Also, more walls means more stretch on top of the part and more compression at the bottom (referencing top and bottom from the test rig).
Annealing (within actual remelting like the salt method) won't fix poor layer bonding. You just need to crank the hotend temperature. Once I started using 250+ for PLA I could never go back. The outcomes are just so much better.
Well, looking at CNC Kitchen and other testing channels, they usually have 5+ of each print and then calculate avarage as well as showing max and min, for each type. I guess that's the way to go :)
This looks like a fun project, welcome to the world of reliability testing and failure analysis.
"We know we will always get what we ask for" is too much to ask from any manufacturer though, you really can't dismiss those parameters from your testing. It helps of course but your test process needs to include incoming quality control to verify things like initial moisture level, filament diameter throughout the spool, etc and tracking temperature, humidity, print speeds and temperatures, etc. I hope that this data will eventually be published, there is a lot of experience in our community and they may be able to help further refine your test process. Definitely increase your sample size though, five is probably the bare minimum I'd personally be comfortable drawing conclusions from.
I'm glad you're talking about at controls and process, you probably should have done that before starting to produce videos. I'm also glad that you did not name the manufacturer, since you've jumped to the conclusion that your data inconsistencies are a flaw in the materials and not a flaw in your process. You said that the material was processed on identical machines with identical settings, but wear and tear and slight differences in calibration, belt tension, etc. can cause variation too. The samples should be processed on the same machine until you can prove that the machines produce the same results on your control materiel.
Until your process is repeatable, if you need to produce content you should make it about your testing methodology. And if you're going to mention a brand, it should be Tangled. There will be people making purchasing decisions based on your tests, and it's entirely possible that you could wrongly destroy someone's livelihood because of bad methodology.
I just wanted to follow up and recommend you get a mechanical engineer to design and maybe perform the mechanical tests. It would give some credibility to your experiments and to this channel. If you’re interested, I’m happy to do a zoom and give pointers on how to design material testing experiments as this is part of what I teach and I am constantly advising my students on materials testing for their capstone senior design projects. This would be pro bono and purely informational (not a substitute for professional engineering advice) to point you in the right direction.
Happy to help,
Lester
Practicing Mechanical Engineer with a math minor here. Is it possible that increasing infill from say, 30% to 40% with the same infill pattern causes where the first line of the infill pattern is drawn on the cross section to shift, thus creating a “different sort of truss structure entirely” on the internal of the part? Of course we know from statics that different configurations of internal members of bridges and such can yield wildly different failure criteria… could the same thing be happening here perhaps? For example, the first “infill line” being started 20mm to the right of the outer left edge of a circular member for 30% infill, but it starting 30 mm to the right of that same outer left edge for 40% infill, just because of how the slicer actually interprets the inputs and slices accordingly.
Or with higher infill layer print time is higher and it cools down more. So much to go wrong. And there is nearly no data to make a conclusion about variance of filament properties. Literally single run :(
what was the sample size per setting? any below n=5 pro is not statistically significant.
I think the sample size was 1......
A lot of people here are suggesting changing only one variable at a time, but that isn't usually best. Focussing on one variable tends to make us ignore other variables that might be important.
Good design of experiment is pretty tough, but totally worth it. If we can think of 6 variables, and we want to try 3 settings of each, and 5 runs of each combination, that's 3^6*5 = 3,645 evperiments you have to run. Good DOE can bring that number down to less than 100 and still provide confidence about which variables matter the most.
increasing walls also increases the amount of time between each layer, a cooler surface doesn't bond as well as a warmer one, PLA is good about just bonding anyways and contact-melting the surface to be adhered to, unlike ABS. But a stress test is going to amplify any imperfections with bonding.
I *LOVE* unexpected data from well-controlled experiments, it means New Knowledge ahead!
Seriously, this is *SO* great! How helpful will it be to know that no, once you’ve hit 30% infill there’s no point in going to 40, it’ll just cost you ire and weigh more but not be any stronger. (Assuming the initial results are correct.)
I can’t say now much I love this, it’s the most exciting thing in 3D printing for me since I got my Bambu X1C! 👍👍👍👍👏👏👏👏
Give it a rest guys! He has done far more than 100000 guys on UA-cam, is opening a second print farm, giving back to the community, running a cheaper filament program, is very entertaining and the totally opposite of a person who is entirely up themselves. Honest enough to put this video out and possibly frustrated at how slowly the tensile testing equipment is coming online. CHILL!!!
You would be surprised at how often the testing equipment itself can give bad results.
I have been called to fix three different online thickness measuring devices because they were all reading off by a similar amount.
Took forever, but I finally convinced the lab, their instrument was off. Not the three online devices.
I'm no engineer, I watch videos and read comments at my best to understand and understanding where this comes handy to my parts and prints, in this funnel I lose a lot compared to who handles the subject for real.
But I come from the Slant3D channel for your great design advices and lemme say dude, starting a channel with a video that talks about your first failure is the bootstamp of someone who's here to stay and stand.
It sounds like the first test for all filaments is the print the same exact thing over and over within at least one, and probably more than one, reel of filament to determine the variability of the filament itself. I was going to say this would be for the cheaper filaments, but in all reality, it needs to be done for every source of filament. I would even go so far as to say that the purchases need to be done over different periods or from different sources to maximize the probability that each individual reel would come from different production runs.
This will establish the confidence level of any other test from a known source with a known variability factor for that manufacturer.
Could the data variation come from the non-linear press movement? Maybe the force should be consistently increasing across the time so that there's no pauses that stress(bend) the samples at different load points to each other? What do you think?
Did you find inconsistency between multiple samples of the same wall thickness? Unless you do that you can't be sure its the material that's inconsistent and not something else going on.
If at all possible, for a baseline understanding, I think it would also be good to compare all these printed results to a solid block of the same material not produced by 3D printing, for example an injection molded block of the same dimensions so you can understand how the material itself behaves with complete uniformity and consistency across the part to compare against the 3D printing process.
I love to see the enthusiasm to quantify and document for us so many material properties related to 3D printed parts and materials, and I absolutely see the vision of how valuable this information will become in reducing so much of the guess work around slicing for 3D printing, just because you've had some good experience with some settings and get a general idea and some "feel goods" about how you choose to slice a part to fit a specific purpose... The analytical measurements of the properties will add so much value. BUT, it doesn't inspire confidence when your entire test fixture is made from the same material with unknown properties that you're trying to quantify, it feels like a hobbiest's approach, print what you can to adapt your tools to the situation, instead of a professional approach where a known and proven test fixture can be used to test your materials. If anything, don't cheap put on your test equipment just to save a buck or a day, both for the communities trust and confidence in your testing results.
I would do some kind of root-cause analysis, like a double fishbone diagram of all of the confounding variables - geometry, print parameters, test setup etc. With so many performance measures and independent variables, I would consider an ANOVA test.
When your Tangled Timeline says change PLA source, does that mean you are moving to a different material entirely, or will you still be producing 850 filament? I am still on my batch of 850 natural/clear, and I am wondering if it will be available in the future (both 850 and clear) as I am seeing significant differences in mechanical part qualities that I like a lot. I really like the consistency of the raw resin and in my experience color always makes for worse consistency.
You should bring an engineer familiar with testing and statistics onboard. That and use standardized tests.
I'm definitely here for the testing, always interesting to see the results of something without having to spend a bunch of time and money on it yourself 👍
I would love to have some videos outlining even just your testing process for each test you plan to use, I believe the collective education of the community here could offer come incredibly valuable feedback to even just testing conditions before you dive right into testing and giving results. Let the community here be your strongest ally as you develop your testing procedures!
Print all the parts with a single print. This way you will eliminate the filament quality variability to minimum since every layer will be printed from almost the same piece of filament. This should get filament problems out of your way. If slicer doesn't allow slicing each part with different settings, you can slice them separately and combine gcode.
This was my thought as well, CNC kitchen does this when he does his testing, all samples of the material in one print.
PrusaSlicer makes it fairly easy to enclose each part on the build plate with its own modifier that allows each part to have unique perimeters, infill pattern and density, top and bottom layers... pretty much anything but changing layer height. All parts on the build plate need to be printed at the same layer height.
@@Liberty4Ever Thanks for the info, I didn't know that, but suspected that such feature have to exist:) the layer height can't be changed because otherwise it would collide with the prints
@@Liberty4Ever I checked it out, PS can now do different layer heights per object. It looks like it slices as the lowest layer height and skips layers until the thicker layer comes.
One thing to keep in mind is as you increase wall thickness or infill density the more heat will tend to build up allowing more crystallization in the PLA is it cools more slowly. The crystalline portion behaves more like an impurity, which could lead to inconsistent test results.
Have you had a yhought about internal stressesin a more dense part??
Could the fact that you’re using 3D printed fixtures and point load have any effect on it? Like maybe if it was metal hardware it would be more rigid and you’d have more consistency across data points?
It makes perfect sense that this linear monotonic progression would be observed in increases in wall thickness with 0% infill, because the overall “truss structure” is the same, with the same stress distribution as the sample before and after it, with the scalar values simply increasing or decreasing. If the fundamental truss structure (i.e.) spacing, geometry, #of members, ect) is changing for every % infill incriment tested, that is from the perspective of the infill, a fundamentally different statics problem with different stress distributions, stress concentrations, and potentially overall different behavior; behavior which may actually lead to a sample with higher infill percentage demonstrating a lower yield strength than one before it.
what was the variance of these measurements at each single configuration? +/- ?
I cant imagine enough to make this much of a difference.
Conflict of interest
I've found the easiest way to eliminate stringing and blobs and etc is to pay $18 for filament instead of $12 or $15. I probably spend more money trying to get cheaper stuff printing than I would just buying a name-brand filament. If I was in Europe, I would definitely buy prusament with all the testing they expose there.
Could you print all infill and wall thickness variations on the same printer at the same time? That would spread the inconsistencies across all parts rather than have a wet or thin section concentrated in only one sample.
I think you are on to something and the first thing that came to my mind was some sort of contamination in the filament. Or maybe its just a cheap filament and they are using recycled material. I did read a paper on filament that shows the more times specific types are heated and cooled as well as how it is heated and cooled can have an effect on the quality of the end product.
2:35
I believe are talking about chain length distribution.
I think you can skip mechanical testing and directly measure the chain length distribution of a sample, the tighter the curve, the higher the control in the manufacturing process (higher quality), of course you can't measure it yourself, you must send it to a lab (Gel permeation chromatography).
When the statistics hits hard...
2:12 I think shows what your issue is:
You are testing with a downward force on a bar supported from one side - ie, a lever! IF you do not place the wedge pressing down on exactly the same spot, or if the bar shifts inside its mount, the potential is there for the amount of leverage to change and the apparent amount of force the bar takes to break changing too.
I suspect your experimental setup is to blame for shifting results and would suggest supporting the bar from TWO sides, to avoid issues of leverage
Having that grid infill inside of a circle is asking for problems where you don't orient it consistently with each test. It would be better to have an outer shape which forces the same orientation each time.
In my limited experience, Hatchbox is good, Overture is literally worse than molten dog-shit.
Doesn't infill orientation play a role ?
You've printed cylindrical test objects that, once picked up from the build plate can be rotated.
Surely when they are then placed on the test bench, with their infills layer lines randomly rotated, that can have an impact, can't it ?
Did you guys take that into account?
One thing I thought might happen is that less walls keeps the object a little more flexible, so it can bend more. More walls result in a stiffer object, which then snaps earlier.
I'm sure you guys are also taking the impact of the part cooling fan into account, and how it might contribute to inconsistency? More infill = longer layer time = more fan cooling etc.
by printing in that orientation you're simply testing layer adhesion, as the walls get thicker the previous layer will have had more time to cool so wont adhere as well counteracting the increase in surface area. By slicing the object in this direction you're not testing the strength of the material or the design. A triangular prism would work well as something easily printable on its side or you could omit the outer walls in a cuboid similar to the wall thickness 'cube' test you did.
I am not sure the infill has to do with the filament, the supports get thinner as you get more so it behaves non liniear
It seems like you're loading speed is too high. Your steps are too big and therefore the force reading might be off. Having recently built a stress testing machine myself I have noticed that consistent in slow paced tests, the best results. Furthermore, plastics have a tendency to creep at constant loading force. It is possible that loading in big steps create a situation where your beam is flexed, but will only break after a few seconds.
all what was tested is layer adhesion. strongest filament i came across is sunlu pla meta @210°C. i doubt your test would manage to break it. sounds funny, but breaking a single wall from flow tests (~0,7mm) is super hard. and you cant go cheaper with filament, it's around 9-11€/1kg
I would have liked to see a scatter graph with a trend line to see just how far each test went besides just the average of each group.
Statistically you should at minimum be printing 9 test samples per test. Really you should do closer to 20-30 for statistical significance
You should really reach out to Stefan at CNC Kitchen to collaborate and learn from his years of experience testing 3d printing materials. His methods may not align to your intent 100%, but it would take a pretty arrogant person to think you can't learn a LOT from his experience. Better yet, I'd be interested to see a proper collaboration video with him at some point to explore what you've learned in conjunction with what he's learned. There seems to be a huge opportunity to share experiences in a way that benefits us all.
I'd love to see a Stefan colab
Awesome!
I have some Zyltech filament which in spots is very brittle. Such a shame, i bought hundreds of Kg this stuff, and their white is the most awesome white i've seen (Ceramic White, truest white i have used)
You are essentially testing layer adhesion, which is analogous to testing for the weakest link in a chain. So you are misusing the numbers you gathered. You think it is a strength number, but it is a weakness number.
yeah please use some quality filament such as polymaker polylite.
These tests seem more for the looks and fun rather than scientific testing, which is alright and has its audience I guess. Just wouldn't draw any big conclusions from the results.
Do some basic analysis to figure out what your minimum sample size will need to be for a confidence interval appropriate for your risk (95%, 99%?). A single element sample size is a terrible idea for numerous reasons and any trained engineer doesn't do it.
Yessss.
Err why aren’t you using your material as a base?
Huhh? I didn’t expect much but I’m still disappointed
A video about a video we didn't make. 🤔
Crying in DIN! 😭
You performed a badly designed experiment and then blamed the material even though you have no proof of that at all. you should have printed a control during all your variations printing. Print the same basic rod throughout your testing and see if it reproduces or not. If it does then your material and process are stable, if it doesnt then you got to solve that problem first.
I came here to roast your ill-conceived test methodologies, overly simplistic understanding of mechanics, statistically insignificant sample sizes, non-existent uncertainty analysis, and completely misguided 'hypothesis' to explain away the 'results' of your 'measurements' and much ado about nothing, but it seems like the other comments here have already stepped up to point these issues out.
There's a reason you don't see bending tests that are configured like those shown here. You also have made a gross oversight by not more carefully considering the anisotropic nature of these 3d prints.
Somebody didn't read the guide @ctrlpew