Quickie link: Turing machine build from wood and scrap metal
Behold! This is the standard table on the Roland MDX-540SA CNC milling machine, costing nearly $37,000. I took a picture of the table and then used Photoshop to eliminate distortion due to perspective and pincushioning. Then I grabbed some guides so that I could measure the positions of the holes.
The Roland specifications show that the holes, which are metric 8mm (M8) tapped, are 110 mm apart in the X axis (horizontal in the picture) and 55 mm apart in the Y axis. There are no specifications whatsoever about the positions of any other holes. This, in case you weren’t paying attention, is a $37,000 machine.
Take a look at the cluster of four holes in the lower left that don’t fit the pattern. Also look at the similar cluster in the lower right, next to the Z-height post. The lower leftmost hole is almost covered by the post. Two out of each of these four holes is a locating hole. But the Roland manuals don’t tell you where they are, or what size they are. I had to figure it out myself, This, by the way, is a $37,000 machine.
Using this photo, and taking a few measurements physically, I was able to determine the exact positions and sizes of these holes. All of the holes are M8 except for four — count ‘em — four locating holes for 6 mm dowels. Four locating holes, two stuffed in the lower left, and two in the lower right. These locating holes are for a 4th axis attachment. So, four locating holes off in the corner of a table on a $37,000 machine.
Did Roland cheap out? Are they acting corporatist? Honestly, how hard and costly could it possibly be to add a bunch of usefully-placed locating holes? I can understand not adding a hole to a part that you make millions per month of, if a hole costs a penny. That’s a savings of $10,000 a month right there.
No wait, I take that back. I never understood that kind of race-to-the-bottom accounting. Does anyone have any idea why Roland would not put locating holes on their table? Please hypothesize in the comments.
So behold, a Google Sketchup drawing of the table (minus the Z-height post, which I didn’t measure). Click here for the Sketchup file. You’re welcome. (Update, 14 March 2011: In fact, the two locator holes on the right aren’t even locator holes. They are slots, and so they are useless for locating. Also see the update at the bottom of this article).
So, we can now begin to design an elevated platform where we can add fixtures. Why do I need an elevated platform? Because the toolhead will only go down to 99 mm (3.9 inches) above the table. This means that if you really, really wanted to crash your tool into the table, you would need to have a tool that stuck out of the toolhead by 3.9 inches. If you’re milling with a 3.9 inch long tool, you must be cutting foam or something. Any harder material and such a long tool would deflect, making a precision CNC machine kind of pointless.
So the end result is, if I want to create a decent platform for my workpiece, assuming about an inch of tool stick-out, I need to raise the table 2.9 inches or so. Furthermore, since my platform is going to be truly useful, unlike Roland’s standard table, it will have many locator holes and many bolt holes that have to be fixed in the same position no matter how many times you take the elevated platform on and off. So the four locator holes on the table are crucial.
Finally, I need to be miserly with the material that goes into the fixture. We have a 1/2″ tooling plate that will serve as the platform top, and it weighs 16.5 pounds. The Roland’s capacity is 44 pounds, leaving 31.5 pounds for the rest of the platform, the workpiece fixture, and the workpiece itself.
The first step is to make two plates, each of which is located by the locating holes on the table and secured with M8 bolts:
The plates are 1/2″ aluminum, which I calculate will weight around 3.8 pounds. I’ll press-fit 6 mm dowel pins into the platform, which will slip-fit into the locator holes on the table. Ideally the locator holes would be more spread out, but thanks to Roland, this is what I’m stuck with (NextFab preferred that I not remove the table, put holes in, and put the table back on).
Next, I will be holding the platform up with pillars made from aluminum tube. So I need to add holes to the plates for attaching the tubes:
Each tube is held in place with two 3/16″ dowel pins press-fit into the plate and the tube, and two screws to make sure the tubes don’t come out.
Next, the tubes themselves go in. How tall should the pillars be? If I want the end of a tool with one-inch stick-out to reach the top of the platform so that I can make fixtures using the Roland itself, the top of the platform must be one inch below the bottom range of the toolhead, which means 2.9 inches above the table surface. Since we have a 1/2″ plate on the bottom, and a 1/2″ tooling plate on the top, that means the pillars must be 1.9 inches tall.
There are two locating pins, thus the need for press-fit holes. Also each pillar has a tapped hole to secure the top plate to.
And that’s it! The tooling plate fits on top, located by the pins, screwed in with four screws.
I’ve ordered the 1/2″ aluminum plate for the bottom, and the tubing, plus various hardware, and some reamers to create the press-fit and slip-fit holes for the pins. Hopefully they will all arrive by Saturday so that I can start building this thing.
Update, 14 March 2011: In fact, the Roland documentation says that origin repeatability when the machine is turned off and on again is +/- 0.0008″. So you’re not guaranteed to find the origin if you need accuracy of 0.001″. Also, over twelve inches, the thing may be off by +/- 0.004″.
The conclusion is that the Roland MDX is unsuitable for CNC. It is marketed as a subtractive rapid prototyping machine, not for CNC manufacturing.
or,
How an Event devastating to Morale resulted in a New Appreciation of Lasers.
PART I:
The Quote Question
So armed with manufacturing diagrams of the parts I needed made, I first paid a visit to mfg.com, which was suggested to me as a good place to find manufacturers who would bid on my parts. After some investigating, I determined that, in fact, the most basic buyer’s account costs US$100 per month, renewed automatically unless you send mfg.com an email. I just couldn’t justify the expenditure.
Instead, I went to thomasregister.com, and looked for CNC-capable small manufacturers. I chose a few from Pennsylvania (i.e. local), and a few from other places around the country. In total, I chose 11 shops. I sent each of them the diagrams, explaining what I needed, and asked for a quote with a deadline of 15 days for a response.
Out of the eleven, three wrote back declining to quote. Another three gave quotes, and the other five never responded. Of the three that quoted, the total prices were $19,000, $30,000, and $101,000. I felt that the $19,000 quote was reasonable, and also came from a shop recommended by a good friend. I let the shop know that they were chosen.
Then I asked them whether they could give me some insight into the costs of their own shop, and maybe the other shops. Why were they a third lower than one shop, and 80% lower than another? A day later, I got the answer: they had just gotten a large two-year contract, and they had to clear their calendar completely for it, which included declining my job.
PART II:
The Morose Maker
I spent the next day moping about. What could I do? I calculated how big a Kickstarter project needed to be assuming that gifts cost 15% of the pledge, fees from Kickstarter and Amazon ate up 8.5%, and federal and state income taxes took away another 20%. The answer was, add 77% to what you need, to get what you really need. So $30,000 would end up requiring $53,000: a huge Kickstarter project. Even if you somehow managed to class your project tax-wise as a business, you’d need to add 63%. I couldn’t even imagine how Ibuilding a steam-powered all-mechanical computer could be justified as a non-profit business (so, no taxes).
So I sat around at NextFab, aimlessly looking at drawings of my parts, turning them upside-down, looking at prototype parts, fitting them together, fiddling with screws, and generally feeling sorry for myself. I went home with no solution in sight. I went to bed, and woke up the next morning with an idea.
PART III:
The Illuminating Insight
Apparently all that fiddling around at NextFab primed my brain to think subconsciously about the problem when I was asleep. This seems to happen often to me: I’ll be sleeping, or taking a shower, generally minding my own business, not thinking about a problem, when all of a sudden a new path of investigation will occur to me. You know how dreams often incorporate thoughts and objects encountered during the previous day? I’m convinced that these flashes require priming the brain so that the problem takes on a sort of prominence in one’s mind. That way, the subconscious gets to chew on the puzzle while the conscious mind does something less stressful for a change.
That’s my theory, and I’m sticking with it.
I woke up with the idea that I should take another look at the casing for each module, and try to make it thicker and out of acrylic. If it were made out of acrylic, I could use the laser cutter to cut the pieces, and not have to rely on an outside shop. My rule is that I can use any material as long as it doesn’t have some essential property that was not available in Babbage’s time. Acrylic is fine, as long as I’m not relying on its transparency or weight for its essential function.
That day at NextFab, I made this on the laser:
This is half-inch acrylic, which seems to have the strength I need. The large notch in the center is new: I determined, using the sample rod that you see above, that the rod would bend under its own weight (plus the weight of the bumps attached to it) if it had to span 16 spaces, but not 8 spaces. So I had to add supports crossing the center. Here’s a drawing I did before I came up with the acrylic solution, showing what I mean.
The perspective kind of makes it look like part of an Aztec pyramid. By looking at this diagram, I discovered that I didn’t need alternating notched and un-notched pieces: I could make a single piece.
Anyway, I started calculating. I can fit about 14 acrylic side pieces on a single 18″ x 18″ x 1/2″ acrylic sheet. That sheet costs $44, so figure $3.15 per piece for materials. I can cut one piece every 4.5 minutes on the laser, or 13 pieces per hour. At $25 per hour, that makes $1.92 per piece for cutting. Then I need to postprocess the pieces in the CNC mill because the laser doesn’t cut nice and straight, and I need to drill holes that the laser can’t cut.
I haven’t tried it yet, but I think postprocessing the pieces will likely take something like 3 minutes per piece, or 20 pieces per hour. At $30 per hour, that makes $1.50 per piece for postprocessing.
The total becomes $6.57 per piece. The cheapest shop wanted $16.87 for the equivalent (one notched and one un-notched piece). I had just cut the largest cost in my project down to 39% of the original cost.
The next largest cost in the project is the rods, and I’m not sure how much lower I can go with that. There’s no way I can do it in-house: the CNC mill doesn’t do steel, I need 34 holes drilled per rod, and I need about 1300 rods.
Part IV:
The Ridiculous Roland
For some reason, the Roland CNC mill is built such that the toolhead cannot reach to within three inches or so of the bed. This means that every piece has to be raised up off the bed quite a bit, which also requires ensuring that whatever is supporting the piece is flat relative to the spindle.
This, to me, is ridiculous.
It means that one has to make some kind of crazy elevated fixture for every piece one mills. So I doodled a bit, and came up with an idea for an elevated tooling plate, atop which one could put custom fixtures, attached to the elevated plate with slip-fit pins. This means that once you determine the machine coordinates for your fixture, you wouldn’t have to zero the machine again if you took the fixture out and put it in again some other day.
My idea requires use of the lathe to create supporting pillars for the plate. The Roland’s bed has only screw holes, and if I screwed pillars in, then the tops of the pillars would change height depending on how tightly they were screwed in. Instead, I figured I would have shorter screw-in pillars, with cylinders that slip-fit over the pillars. This way, it didn’t matter how tightly the pillars were screwed in: the cylinders would determine the height. Then, pin a flat plate on top of the cylinders, and we have an elevated tooling plate.
Alas: the lathe at NextFab somehow got its spindle out of center, meaning that I can’t make round things. They come out like ellipses. The repair guy is due to show up some time in the next few weeks :/
Part V:
In Conclusion
So you probably won’t see much progress on the Engine over the next few weeks. Making the tooling plate and the fixtures will take up a lot of time, and once that is done, I will slowly start working on making the casing pieces. But I will blog my progress on the tooling plate.
I made these steampunk thumb drives out of Kingston 4GB DataTraveler thumb drives. They’re for sale at USD 25.00 each on etsy.com. The money goes towards buildling the Logical Engine!
I only made five, but if they go quickly, I can make more. Go take a look!
Got bored, so I played around with the book cover template in InDesign. What if Charles Babbage and George Boole had brainstormed together? And what if Babbage ditched decimal representation in favor of binary? And then went on to build the much simpler and cheaper Logical Engine? I think this would make a fun alternate history, one where *I* didn’t have to build the damned thing because Babbage already did it 150 years earlier!
Image of Difference Engine is cc-by-sa by Flickr user Gastev.
Tags: babbage, boole, logical engine
A few months ago I took a trip to see my good friend Dan Reetz in Los Angeles. He took me to Norton Sales in North Hollywood, a surplus store specializing in space program castoffs, and which has an enormous warehouse of same. We spent a happy hour clambering through the mostly unidentifiable technical detritus of the nation’s space program.
I found some welding goggles and mugged for the camera. Thanks to Dan Reetz for the image and the post-shooping. I don’t know what that thing that I’m holding is.
I found a Univac tape used in the Polaris program. There are fun things like this tucked into all sorts of unlikely corners.
Also, an individually-wrapped screw. Your tax dollars at work.
Here are some of the promised drawings, which I drew in SolidWorks, and exported to PDF format. I’ve sent these out to eleven machine shops around the country, asking for quotes.
I tried using emachineshop.com’s CAD program, which promises an instant quote based on your drawing, but for 1400 rods (each of which is a 14.5″ x 1/4″ x 1/8″ steel bar with 36 holes drilled in), the instant quote came back as around a third of a million dollars. Clearly something is wrong, but I’m not in the mood to fight with their program to figure it out.
Anyway, required for the ALU are (click on each part to get the PDF):
- 1242x Rod R16
- 444x Notched 16
- 612x Blank 16
- 280x Drive Bar 16
- 280x Drive Arm
This doesn’t include the helical drive, bearings, springs, or bumps, which I can buy or machine myself.
I found that adding a key to the Acme rod and sliding a chain sprocket on acted as a perfect driver for the module. The key allows the sprocket to slide as the Acme rod moves up and down, and can rotate the rod at the same time. The sprocket can be driven by standard motorcycle chain.
Since that is the final design challenge, I am ready to start getting quotes for parts, but first I have to draft drawings of the parts. I’ll be posting the drawings as I complete them.
Based on the plan in the previous post, I have completed and tested the spiral cam drive. First, I cut to size some aluminum round stock on the lathe, and drilled a 3/4-inch hole through, also on the lathe, for the Acme rod. This became my blank.
Next, I wrote a program to generate GCode to mill the spiral cam out of the blank. I went through several iterations, improving the efficiency each time. It still takes about 20 minutes to mill the spiral cam. And here is the inevitable CNC porn:
I drilled and tapped a hole radially into the cam, drilled a set hole radially into the Acme rod, and attached the cam to the rod. Here’s the installation showing how the bearing rides on the cam surface:
And here are some videos showing the mechanism in action, driving 12 rods. Four of the rods have extra-strength springs so this is the equivalent of 16 rods. The action is incredibly smooth, and doesn’t take much force.
I also tested stringing fishing line between two modules, and that worked perfectly. I still need to work on getting the length of the string just right, since I can’t adjust the spacing between modules for a given string length.
NextFab is closed for two weeks for the winter solstice holidays, and this also ends this phase of investigation. The next phase, to be started in January, is making a cylindrical gear to rotate the drive, and then putting together a single bit slice module. When those tasks are completed, I will have enough proof that I can complete the 16-bit ALU, which means that I can start a Kickstarter project. I hope to complete the bit slice module by the end of February.
Here’s video showing the new drive. It’s sooooo much easier to move. Next will be attaching a bearing to the drive bar, and milling a fusee to push against the bearing. I have high hopes for this design!
Improved 16-rod drive test from Robert Baruch on Vimeo.
Twitter
RSS