Week 10: Machine Team Project

This week was the team machine-building project, and I was pleased to be part of the highly competent and diverse CBA section "Makey Cakey." We started by deciding upon what to make. I gave up on the chocolate printer, since doing a full 3-axis machine was going to be way too tricky. My team was excited about food, though, so we quickly agreed upon a pancake printer that would be way cooler than PancakeBot . Then we can put it up by the Media Lab's Foodcam so everyone can print their own customized pancakes!

We divided up into three teams: software, hardware, and mechanical engineering. I joined the engineering team, which then split up further into the end effector group and the base machine group. I wound up in the base group, and we planned to meet up the next day to CAD but a few team members went ahead and completely modified Jake's machine design-- so our meeting turned into a ShopBot setup session rather than a CAD session. Which turned out to be perfect, because what appeared to be simple milling tasks on the ShopBot transformed into a series of epic endmill-destroying disasters.

I helped Yun out a bit with the 3D printer, which ran out of filament mid-print and then some of the filament got stuck up inside and we had to check the Wox website to troubleshoot. My main contribution, though, consisted of controlling, supervising, and troubleshooting the ShotBot from Friday through Sunday (or early Monday morning, to be precise). The whole sordid tale is on our team's documentation page, but here I've copy-pasted what happened in my subsection.

CAD --> CNC Milling

Preparing and then milling out the toolpath was far more exciting than we'd anticipated. First Oscar and George spent hours modifying Jake's machine to fit our griddle, and then unfolding it in Fusion to create the layout of parts for milling. Then they painstakingly created a bunch of CAM toolpaths, which involved clicking in the middle of hundreds of screw-holes on our model to ensure that the holes all got milled before the main pieces-- only to find that we could just use Jake's original toolpaths for the most complex framing pieces. We couldn't figure out how Jake made all the sketch-based toolpaths for the 45˚ angles, so it was an easy fix for Anna and Burhan to just copy-paste Oscar's modified parts into Jake's more complicated toolpaths.

There were nine different toolpaths to mill, involving three different endmills-- because Jake is a freaking genius who foresees everything and carefully optimized the whole milling process to mill the right things in the right order. Originally he even had four endmills, but realized that he didn't need to be quite so precise about optimizing the right tools for the right jobs.

Burhan, George, Anna, and Vik then had a series of ~~disasters~~ learning experiences while attempting to mill out our model:

First of all, Fusion and the Shopbot software both had some sort of grudge against us and kept freezing at inopportune times throughout the day.

Since we only had one piece of HDPE, we sensibly ran an air-run before actually milling-- and discovered that we'd forgotten to zero our axes on the right place in our CAD model. We fixed this, and also moved one of the pieces that was a bit too close to another piece to give it some better clearance.

As we were zeroing the Shopbot axes, the milling head kept running into its own vacuum tube and bumping backwards and then freezing the software. We had to pull the vacuum tube up through its hooks on the ceiling before we could properly zero X and Y on the corner of our material.

The first toolpath (milling out the Z-axis mounts on the complex pieces) milled out smoothly. Then we changed the bit to the 1/8" that we would use for toolpaths 2-8... and the bit snapped in half after a few minutes of milling. Probably we didn't tighten it on enough. Luckily Jake was at the lab on Saturday, and he came to bail us out with a spare 1/8" bit.

Our next toolpaths were supposed to bore all the holes and fingers, but the mill only cut clear through the material on Jake's original pieces! What had we done wrong? Jake had to come save us again, and he pointed out that we'd chosen our toolpaths on the backside of our CAD model within Fusion-- so we had to go through and re-select all the holes on the front side of the model and make sure that the milling arrows were pointed the right direction.

We loaded up the new toolpaths, and the Shopbot finally started cutting through... but all the new holes were off from the original milled holes. We zeroed our X and Y axes on Fusion and tried again, and then they were only slightly off. Had the Shopbot somehow lost our original zeroed axes? This was possible, since the Shopbot software had kept freezing. So we tried manually offsetting the axes, but the problem continued...

... and it turned out that Anna had loaded the wrong version of our CAD model, before we'd moved the pieces around to ensure clearance. Once we sorted this out and re-zeroed our Shopbot axes, the hole-drilling toolpaths finally milled out properly. We marked the wrong holes, and hopefully all those extra holes won't be a problem.

Then, when it was finally time to cut out the contours, we ran into the same shallow-cutting issue on the new pieces! Argh. Back to the Fusion drawing board.

Burhan sorted out all the contours and we milled almost all the new pieces without a hitch, and then one of the smaller pieces hasn't been screwed down properly. It flew up dramatically and broke the endmill-- the last 1/8" endmill left in all of CBA! All was lost...

...until we begged a new not-quite-1/8" endmill off of the EECS lab, and then we managed to break that one too. We're still not entirely sure what happened, but it might have something to do with the fact that this third endmill ended in a long skinny bit and didn't have all the little grooves that our original endmills had. Anyway, it got a bunch of little plastic bits all stuck up and then it snapped. We're sorry, EECS.

It was now 8pm on Sunday, and we called upon Jake for his advice. Lo and behold, he had another 1/8" endmill stashed away! This was really the last one, so we promised to be careful. Erring on the side of caution, we stopped milling after a few minutes because all of the plastic bits were getting stuck in the groove. This hadn't happened before, so we ran up to ask Jake for his advice. Poor Jake was still in his lab doing important circuit-related things, but he suggested tinkering a bit with the spindle speed.

Terrified of breaking our last endmill, we decreased the spindle speed by a moderate 500rpm. We loaded up the file, started the ShopBot, and then broke our 4th %&$#@! endmill! It got all stuck up in the plastic shavings and snapped. But why did the ShopBot hate us so much? We despaired, shutting down the ShopBot, bemoaning our fate, and resigning ourselves to the misery of life without a functional pancake robot...

...but Jake had a super secret fifth 1/8" endmill hidden somewhere in the bowels of CBA! One last chance! So we reloaded our run, and it finally milled out perfectly! This was pretty strange-- what had cursed our Shopbot last time? Maybe we should've just restarted it instead of toying with the spindle speed?

Lessons for the Future

I then skipped out on the rest of machine-building to catch up on life on Monday and Tuesday, and returned to the shop Tuesday night for the final preparations. Here are my personal takeaways from our weekend of ShopBot woes:

Things can go wrong. Be prepared to go back to the original CAD file and redo toolpaths.
Have one person in charge of the entire milling job, to facilitate change-overs when people need to sleep and to ensure that everyone knows where the most up-to-date CAD files live.
Very hard plastic is hard to mill! Have the right bit and the right settings.
Lower the RPM very slowly-- don't lower it by 500 all at once and expect things to be better. Maybe you can raise the RPM when things aren't milling out right.
Make sure that all pieces are properly screwed down or have tabs! Even when they don't fly up and destroy your endmill, untethered pieces tend to slide around during milling and can mess up your machine's precision.
The ShopBot has something against Burhan, who's otherwise a great guy.
When the ShopBot is acting moody (i.e. not milling out as smoothly as before), first shut it off and on again.
The ShopBot software hates everybody.
Make sure you're selecting toolpaths on Fusion on the top layer, not on the bottom layer.
The more milling programs you have, the easier it is to start over from the beginning of a program when something fails.
It's not always email to just send the bits around and expect people to duplicate your project exactly, especially when there are lots of subsystems. There's a lot of tacit knowledge that's very hard to document and share.
Jake is really incredibly talented. Thanks for all your help and brilliance, Jake!

By the time class started on Wednesday, everything was assembled and working independently-- but some issue with the TinyG confusing millimeters and inches made the Makey Cakey run erratically rather than actually print any coherent shapes. This was disappointing but honestly not unexpected: one week isn't nearly enough to build a complex multifaceted machine, unless you're Jake, Jens, or Nadya. But hey, at least we know what's wrong and plan to fix it in the future!