-
I got the calibration tools from the guy who used to support these machines, and did the on-head alignment camera calibration, and then did 14 more boards, tweaking the camera alignment offsets until the placement of the QFP-44 chips were as good as I could get them. I would guess they were coming in with +/- 0.05 mm accuracy, or maybe just a little worse. Plenty good for a 0.8mm lead pitch. I will have to see how well it does on higher density parts.
I had a few feeder issues, I will have to go through these feeders and deal with rollers that don't grip anymore, etc. But, I'm getting the process down pretty well. The tray holder and setup for waffle trays works like a charm, now that I understand how to set it up.
Jon -
I am jumping the wagon a bit late, but could not resist - your painfull path seems so similar to mine. Only I am dealing with Universal GSM's. However, it was well worth the effort. Alhough machines
I have run on OS/2 (yes, OS/2), software is actually pretty good. Most of the software issues you faced simply do not exist on GSM. ASCII file importer is extremely flexible and you can set-up
the board in a few minutes, if the parts used are already in machine's library. Also, UIC has pretty good web site with documentation and even some troubleshooting advices. They require
500$ sw. registration fee if you bought used machine, but it's well worth since you get access to that web site. I am running machines with 4-spindle Flex heads and now I am almost done with
bringing-up two-beam machine with 7-spindle heads. A lot of joy... -
I am jumping the wagon a bit late, but could not resist - your painfull path seems so similar to mine. Only I am dealing with Universal GSM's. However, it was well worth the effort. Alhough machines
The original hard drive died, somebody put in a replacement drive, which is ALSO 20 years old, and was not even bolted into the machine, just hanging by the cables! I was moderately amazed when the machine booted up! But, I think there could be software rot or a silently damaged sector there. One of the data import modes loses rotation data, and the "optimizer" that is supposed to order the placements for fastest assembly of a board totally scrambles the placement file. It is like a spreadsheet where it sorts on one column without moving the rest of the row with it.
I have run on OS/2 (yes, OS/2), software is actually pretty good. Most of the software issues you faced simply do not exist on GSM. ASCII file importer is extremely flexible and you can set-up
the board in a few minutes, if the parts used are already in machine's library. Also, UIC has pretty good web site with documentation and even some troubleshooting advices. They require
500$ sw. registration fee if you bought used machine, but it's well worth since you get access to that web site. I am running machines with 4-spindle Flex heads and now I am almost done with
bringing-up two-beam machine with 7-spindle heads. A lot of joy...
At some point, I may try to build a new software system, with a new(er) OS and see if I can get a complete set of files for the machine software.
I did manage to clone the old, but working, hard drive to a new drive, and there were no errors reported in the process.
I am still learning how to use it, but have gotten two batches of boards done, so that is programs for 4 sides. It is getting easier. I am setting up my own parts library, and entering my own vision alignment settings, and they work! Without their optimizer, which really is supposed to do a lot of the data import task, setting up is quite cumbersome. And, I'm still dealing with some flaky feeders, and learning the ins and outs of those, too.
Jon -
Well, I'm a convert!!! I was a bit skeptical about the Quad feeders. I have extensive experience with the old Yamaha/Philips feeders on my old CSM84. They needed a LOT of babysitting, as the cover tape peeling was marginal. Also, some tall parts in plastic tape would lift the hold-down plate and the sprocket would jump teeth.
Well, I had a few issues with the Quad feeders on my last batch of a board run, and fiddled with them a bit. I just ran 9 boards with 292 parts/board, and had ZERO pickup errors for the entire run. That's a total of almost 3000 parts. So, I am now a believer that these Quad feeders are a huge improvement over the Yamaha all-mechanical feeders that advance ther tape on the return stroke of the plunger that actuates them.
The only issue I've seen with the Quad feeders is if the tip of the plastic tape catches on an edge in the track inside the feeder, the drive sprocket chews the tape up pretty badly. Maybe snipping the corner of the tape would prevent it from catching, I will have to try that.
As for speed, the machine did 292 parts in 4 minutes 17 seconds, so that is about 4100 parts/hour. Plenty good enough for me!
As for the "bit rot" that I was worried about, it seems that if you set everything up properly (nozzle assignment, part profiles, feeders, etc. ) before running the optimizer, it DOES work correctly.
Jon -
Great news Jon and congratulations! I can imagine the satisfaction in getting to this point!
Interesting to hear about the quad feeders. The Yamaha CL feeders, while ubiquitous, do have their quirks and over the past 6 months we've learned all of the tricks to help ensure they run smoothly. As you mention, the biggest issue is the various conditions which allow the sprocket to misalign.
4100 CPH is a nice speed. It's the average of what we achieve on our new Kayo 6 head machine and while of course faster is always better it's proving to be an adequate speed. -
Great news Jon and congratulations! I can imagine the satisfaction in getting to this point!
Yup, the problem I had with the optimizer was just doing things in the wrong order. You have to define all the parts and vision profiles FIRST (should have been obvious, really!) before running the optimizer, since that searches the part database to link placements on the board with components and available nozzles. When it is done in the right order, then the optimizer assigns heads and nozzles to the parts and organizes the sequence.
Interesting to hear about the quad feeders. The Yamaha CL feeders, while ubiquitous, do have their quirks and over the past 6 months we've learned all of the tricks to help ensure they run smoothly. As you mention, the biggest issue is the various conditions which allow the sprocket to misalign.
4100 CPH is a nice speed. It's the average of what we achieve on our new Kayo 6 head machine and while of course faster is always better it's proving to be an adequate speed.
Then, I had to replace the peel rollers on the feeders to get good pull of the cover tapes. After dealing with those issues, it all really started to run well!
It took 6 years or so to get really proficient with my Philips machine, so I'm doing pretty well already. As for the speed, I have been running the machine at 60%, that seems to be some kind of default setting. I have not played with that, but sometime I'll try turning it up just to see what it does. But, I may keep it at 60% just to save wear and tear on the gantry.
Jon -
Let's see how long before you crank up the speed further. I know I wouldn't be able to help myself
-
Placement speed is a curious fish and the such a moveable metric that promo literature and how it is measured is full of cheating even when they try and apply standards to it. I haven't fully optimised the feeder location on the current run on our Essemtec and the screen currently has a live cph on screen of 5400, this will drop off radically as it starts to place things like ICs at the end of the panel and so uses fewer nozzles and slower feeders. It's also a lie because it doesn't start that count until its located the PCB and loaded its first 4 nozzles, and then over a run you have the overhead of manually securing each panel as its standalone. Even with a very densely populated board and parts loaded in feeders very close to the camera, I have never managed to push that number past 6200 and yet this machine has a theoretical speed of 15k and an IPC one of somewhere between 9k and 10k. I lose up to 25% by not having the fastest feeders, I will grant you that but that still leaves me some way off even its "IPC" rate in a best case scenario.
Luckily 4-6k is still a really good speed for a lot of product types and certainly radically faster than conventional assembly, where it doesn't work is for any kind of volume where a client might start expecting to see a low price. Its also a window a lot of machines seem to operate in and was IIRC a number a lot of entry level commercial machines hit in the real world about 10 years ago. These days we seem to have 6heads plus even on the entry level specs and a headline rate about 2-4 times higher than it was back then.
Interesting that 60% seems to be some kind of default, I'd be suprised if cranking it up caused any issues unless there was a physical issue it was hiding. -
Yup, my Quad QSA-30A has an advertised speed of 17K PPH, which now that I've seen it run is obviously a prepared special case to get the highest number possible. (I have seen videos of some insane machines with a wheel of nozzles. It goes to the feeder area and picks up 20 parts, a camera at the top of the wheel checks the rotation and centering of the parts, and then it goes to the board and places the 20 parts. it places 20 parts in about 2-3 seconds, from pick-up to
placement. That was QUITE impressive, but such a machine is a chip shooter, and can't handle a wide range of part sizes.)
Anyway, my guess is turning up the speed on a small board with all the feeders next to it will have very little effect on speed. It is all in the acceleration. If you are placing huge panels with feeders far from the board, then it might make a difference.
Jon -
Literally a wheel is a Universal Lightning head I believe and its even 30 parts at a time, you can have a machine with one of those and a 7 at a time flex head tho http://www.uic.com/wp-content/uploads/2014/01/Fuzion_Platform_brochure.pdf. But its just as big and expensive as two machines...
Plenty of others have a rotary turret ASM and Europlacer manage to do that and be flexible, although its 12parts at a time and a fair bit slower (still impressive to watch). Interestingly when Europlacer targeted higher speeds they dropped the turrets and went back to a row of nozzles, but even they admit its a compromise as each head is limited to 1/4 of the machine so you have more work to balance the parts.
The two sides seem to have endless arguments about which is better, a line of nozzles can gang pick but that in itself makes some huge assumptions about how well you can optimise your loadout or indeed how many reels of the same thing you want to buy. -
So, the QSA-30 is now my only P&P machine (hope it doesn't get lonely.) I sold my old Philips CSM84 to another eevblog member and he picked it up last week. Lots of ruts in my back yard, but we got it out with some help from towing straps and 4WD trucks.
And, I now have a bunch more space in my shop.
Jon -
That CSM84 sure have 9 life just like a cat!
-
That CSM84 sure have 9 life just like a cat!
Well, I think all the old-school P&P that were very well made just keep going. The Yamaha/Philips CSM series were built like tanks, and used very rugged consumables. I made my own nozzles on my lathe and mill, and made a few repiar parts when I broke the ends off the air pistons that drive the nozzles up and down.
I ran that machine sparsely over the 13 years I had it, but did make over 2000 boards on it. In that time, I had a sensor on the conveyor fail and the commutator of the rotation motor filled up with copper dust and started giving errors. That is the entire number of issues I had with the machine. Like I said, built like a tank!
The Quad QSA30 is also quite robust. I don't have long experience with it, it is WAY more complicated (8 motors, 4 cameras, 14 computers, etc.) but seems to have a lot of sophistication and self-checking built in. Now that I've experienced the Quad, I can see a lot
of things that the Philips kept hidden. For instance, when the Philips located fiducials, there was no way to know the quality of the location or what offset it was applying to the board origin. There is a test fiducial function on the Quad where it shows you the video with computer-drawn crosshairs so you can see that it has correctly registered the centroid and then gives the offset it will apply.
You can also do a "test pick" and it will pick a part from a feeder and report the X and Y size and X, Y and rotation alignment of the part. This tells you if it is picking up the part from the center of the tape pocket. Since centering on the Philips was with mechanical jaws, no such readout was possible. Recent experience shows the Quad feeders are much better at peeling cover tape than the Yamaha feeders, that was one constant issue that the feeders needed babysitting. But, the real reason I moved up from the Philips to the Quad was the QuadAlign camera system that uses line-scan cameras to center the components. The Philips centering jaws have a certain amount of unrepeatability, and seem to only handle parts with a certain aspect ratio. So, oblong parts like resistors and caps and SOIC chips were fine, but a square chip like a QFP or MSOP would not be aligned in both axes. the last board I did on the Philips had 13 parts I had to nudge into proper alignment under a microscope. This pushed me over the edge.
Jon -
Well, the saga continues. There is something fouled up in the configuration files on this machine. There is a menu to set up the Automatic Nozzle Changer, you can set the location and what is in the nozzle pockets, and pick and put nozzles in and out of the
pockets just fine. But, when the machine goes to assemble a board, it goes to some preprogrammed position that I have no idea
where it comes from. VERY frustrating. Also, the machine doesn't seem to like to have the nozzles set up on the head by hand, it wants to pick them up itself. I have not found the config file that holds the ANC pocket positions. It took me TWO WHOLE DAYS fooling around to get it to where it would accept preloaded nozzles and just use them as is.
Jon -
And, I may have cracked this issue. I had been copying my old pcb files to make new ones, and that apparently copies the ANC settings.
I created a new, empty pcb file and set up everything, ANC, feeders, parts, fiducials and placements, and it tried to grab the nozzles from
the right place. But, maybe the alignment wasn't perfect, and the nozzles jammed. But, I think if I just adjust the coordinates of the
ANC pockets a little, it is going to work. Having automatic nozzle change is going to expand what I can do. A board I need to run
in the near future has everything from 0603 to 20mm FPGAs on it. That is going to require the ANC.
So, I was able to go from no setup at all to running 17 panels of boards in about 6 hours. That was a big improvement from running into
all sorts of issues getting the setup to be accepted on the last board.
Jon -
So, I was able to go from no setup at all to running 17 panels of boards in about 6 hours. That was a big improvement from running into
all sorts of issues getting the setup to be accepted on the last board.
Wow, well done! Great to read stories like this.
I've fixed up a few CNC machines, but I'm unlikely to take on a PnP requiring so much effort.
-
Wow, well done! Great to read stories like this.
Well, yes! Had I known that the machine had not been turned on for about 6 years before it came up for bid, or that it might have been stored in unconditioned space in Austin, TX for that time, I might not have placed my bid. I DID place a pretty rock-bottom bid, and got it for the opening bid price, ie. NOBODY else bid on it! That left me with a lot of budgeted $ to spend on repairs and accessories.
I've fixed up a few CNC machines, but I'm unlikely to take on a PnP requiring so much effort.
I was VERY comfortable with my old Philips CSM84, knew it inside and out, but it was a very simple machine without ANE, Z axis or vision. I had very few issues with it over 13 years. Well, little did I know how messy things could get, especially with a much more complex machine. The Philips had ONE CPU to run the whole thing, and no OS. The Quad/Samsung has, I think 14 CPUs of various types, and tons of software.
I think this machine was the best one of the low-end machines in that auction, careful examination of the photos indicated that a number of the other machines had been partially disassembled and had parts missing.
So, I have now done 4 different boards on it, 3 of those were double-sided. It is going to be a while before I am as comfortable with it as I was with the old Philips, but I will get there. This machine is so much more sophisticated than the Philips. I have yet to actually use the automatic nozzle changing. The vision alignment and programmable Z is a big plus. And, the ability to make it show me the results of the fiducial pickup and test pick alignment is very nice. Also, I now know how to do array boards (panels). The Philips could do it, but I never learned how.
Jon -
I got this machine in August, 2020. I never got the ANC to work. One of the issues is you can set up the ANC and teach the hole locations in a GUI, and it all works perfectly. But, when you go to place parts on a board, it uses some OTHER data, hidden SOMEWHERE, for the ANC coordinates. I finally figured out a sequence to get that to work, but over several months, I have NEVER figured out how to change which air and sensor port the ANC is connected to!
So, I was forced to put the ANC back on the front feeder rail and hook it back to port #1. Now, the ANC acutally tries to work. But, head 3 jams instead on nicely putting or picking the nozzles. More investigation shows the "Z-rod" on head 3 is bent. Initially, I measured the TIR runout at .015", with some struggling I have got it down to .006". I think I'll try to work on it some more. Head 1 has less than .003" TIR. That would explain ANC issues as well as being unable to pick up an 0603 part. It would come out on the side of the nozzle.
Fixing that runout and recalibrating the head offset should help things a lot.
Well, it was really good to see the ANC work like it is supposed to. I made up a test program mounting 3 parts that needed a different nozzle for each. It didn't work on head 3 due to the runout, but when I reset it to use head 2, it worked perfectly! The top side of the board I'm working on now will have everything from 0603 to a 20 mm FPGA chip, so getting the ANC working was necessary.
Jon -
Well, I had to use shade tree mechanic skills on a $100K machine. Last year I replaced the #3 head with an eBay head, due to the trashed ball spline on the head on the machine. It all seemed to be working OK with 0805 parts, but i was
having issues with 0603's. So, I had to apply massive force to bend the Z rod closer to straight. I managed to do it without breaking anything. I got the TIR runout down to .003", which was only a little worse than the other heads.
Then, it wouldn't pick up sonme 0603 caps. I did notice those caps were a bit thinner than the tape, so they sat BELOW the surface of the tape by a few thousandths. Also, when it DID manage to pick a part, it would blow it all over the board, meaning it wasn't planting the part. So, I tried changing the Z offset on that head, and now it picks and places the 0603's fine!
Also, while dealing with this yesterday, I got some dirt on the alignment camera, and it was giving crazy dimensions of the parts on that head. I had to pull the camera to clean it properly. Once I could SEE the speck of dirt, I could clean it off. (The software has a diagnostic page where it gives you what looks like a scope trace with X as linear dimension and Y as brightness of the linescan image. So, you can easily see when there's something wrong with the camera.)
Well, I hope this is the last of these issues for a WHILE!
Jon -
Awesome work Jon, I really enjoy reading about this process and understanding the differences between these highly engineered "classical" machines and our modern Chinese "economy" machines
-
Awesome work Jon, I really enjoy reading about this process and understanding the differences between these highly engineered "classical" machines and our modern Chinese "economy" machines
Yeah, this has been quite a saga! The machine sold at auction for $500, but I've had to throw in an addtional $7000 or so for repair parts and then feeders, a replacement head and a few other things. I built my own waffle tray holder.
I've now completed 4 production runs with it, each time going to a more challenging board. Now that I have the ANC working, I can move up to more challenging stuff. I'm doing the back side of another board now, then I will do the front side, which is about as difficult as anything I expect to do. It has everything from 0603 passives up to 144-pin 20mm TQFP FPGAs with 0.5mm lead pitch, and MSOP10 chips. These are the ones that started the whole process of upgrading. My former Philips CSM84 could do a fair job with the big FPGA, but the centering jaws just did not work AT ALL on the MSOP10! It would put them randomly a whole lead pitch off, and there are 8 of them per board. A total of 14 ICs that had to be manually poked into alignment. And, of course, if you fail to properly realign one, it is a pretty big hassle to fix it after reflow.
Although my Philips could do panels, I never tried it, as there was no way to know if it was set up right.
The Quad has a test feature for nearly everything. Once you have entered the array X and Y spacing, number of repeats and the fiducials, You can load a board, and walk it through each board of the panel and view the (0,0) coordinate on the camera.
The QSA30 has actually quite nice software, once you understand it. The manuals are about the worst I've ever seen.
So, after 4 failed attempt so pick a part, it marks that feeder as skipped with a message, and then goes on to complete the rest of the board or panel. Then, at the end, you go in and fix whatever was wrong, like cover tape not peeling, and press start. It then does only the missing parts -- pure MAGIC!
It also has LOTS of online diagnostics, which has been a huge help. So, you can select a feeder and do a "test pick".
It picks the part, measures it with the alignment camera and gives a dialog box with the X and Y correction required, as well as the X and Y size of the part. So, you can verify the alignment camera is properly analyzing the part.
You can also view what the flying alignment sees at that point, so it gives you a "live image" of the part on the nozzle.
This flying vision is a line-scan camera at each nozzle, mounted on the head, so it gauges the XY size and alignment correction while the head travels to the board. You can also pick up and put away nozzles in the ANC, or tell the machine that you have manually placed a nozzle on a head. You can teach the position of the ANC pockets with the down-looking camera. You can also program the location to pick up a part from a feeder if it is not at the normal position, using the camera. As for fiducial recognition, you program the location and size/shape of the fiducials, load a board and you can then tell it to scan the fiducials. It shows the camera image on the screen and a black crosshair where it thinks the centroid is. So, you can tell if it is properly doing all this. This was invaluable when I was diagnosing missing solder joints in the DSP board. And, when I fixed that, it was immediately CLEAR that I had, in fact, fixed the problem.
It took some time to get it to work, but I found a way to download a "CAD import" file and then run the "optimizer" which figures out the best place for feeders and the best order to place parts, optimizing for head motion and nozzle changes. I wrote a C program that takes in the P&P file from my CAD package and outputs what the QSA needs.
So, this machine has X and Y servos, 3 steppers for head rotation and 3 servos for Z position. It has 4 cameras, 3 of the line-scan alignment type and one down-looking for teach and fiducial pickup. It has a PC for the GUI and program loading, a Motorola 68040 VME computer to run the motion system, and a total of about 14 computers, all the rest are single-chip micros or TI DSP processors.
This machine is VASTLY more complicated than my old Philips, or the Chinese machines. In 2000, they needed a distributed system to handle the workload, all the vision stuff going on, etc. But, I'm sure glad they did put in all that diagnostic and test functionality, it makes it a lot easier to be sure you are setting up the program right.
When I first got it, I thought the Quad electronic feeders were over the top complicated, and would be a lot of hassle to use, compared to the totally mechanical feeders on my old Philips. But, after running a couple boards, I see how good they are. The Philips/Yamaha feeders had LOTS of issues, mostly peeling cover tape, and the tape binding under the hold-down plate. Those didn't have a latch to keep the holdd-wn plate from rising up, and the advance was made by a spring after the head released the trip lever. So, I had to hover over the machine waiting for the inevitable feeder jam every couple minutes.
The Quad feeders are not quite so easy to thread, but since I replaced the rubbery peel rollers, they are VASTLY more reliable. I did a batch of boards in 2 runs, placing about 2600 parts with ZERO feeder errors in each run!
They also have electronically programmable pick-up position and parts pitch, that you set by pushing buttons.
Jon -
Luckily 4-6k is still a really good speed for a lot of product types and certainly radically faster than conventional assembly,
In my world 4-6k is terrible. I went through alot of pick place machines before getting my Fuji Turret head. Everyone gave me a lot of jive about cph and unless its a turret head it wont come close to rated figure. Turret heads are generally within 10% of rated figures ~36k cph. Since then to now I have been very anti piano head as I like to call them. Supposedly the old phillips multiflex/assembleon ax series can do high rated cph figures. I was trying to find an old phillips to play with. My Fujis CP6's have been great. I wish I hadn't listened to everybody who pushed gsms/jukis/etc. Wasted alot of my time and energy trying to get near IPC cph figures. I do like Samsung machines as they seem pretty straight forward. -
Well, I'm not in that world. I make batches of 20 - 60 boards at a time, usually. I have a converted toaster oven that can take 4 - 8 boards at a time. So, really, 4K CPH is pretty good, and can keep ahead of manual stencil printing and the batch oven.
Luckily 4-6k is still a really good speed for a lot of product types and certainly radically faster than conventional assembly,
In my world 4-6k is terrible.
Jon -
Sorry I didn't mean to come off patronizing, just trying to contribute to discussion. I love pick place machines but have a hard time finding information about each brand/model without buying them and playing with them. I was trying to complain how CPH figures are seriously misleading especially with piano style head.
Very impressed with your Quad story. More because the model you got was a samsung. Somehow Samsung made units easier to work with 20 years later and easy to keep alive. -
Sorry I didn't mean to come off patronizing, just trying to contribute to discussion. I love pick place machines but have a hard time finding information about each brand/model without buying them and playing with them. I was trying to complain how CPH figures are seriously misleading especially with piano style head.
Yes, this machine has a 17,000 CPH rating. Clearly a bunch of baloney on a job specially designed only to give the highest number possible. They do have synchro-pick, where the spacing of the heads is a multiple of the feeder rail spacing, so if the parts are put in the right feeder slots, it can pick 3 parts at the same time. I've set up a job to use that feature. It really doesn't speed thing up by a large margin. Oh, and for now, I'm running the machine at the defauly 60% rate. Sometime, I will have to see what it does at 100%.QuoteVery impressed with your Quad story. More because the model you got was a samsung. Somehow Samsung made units easier to work with 20 years later and easy to keep alive.
Well, I did play with a Samsung CP45 with the mirror-flip bottom camera scheme at an auction. It looked quite nice, but was way more machine than I needed. Since the CP30/QSA30 is pretty old, now, the vultures will let parts go for cheap!
The manuals are absolute CRAP. Clearly, Quad just took the Samsung manual and edited in the pages relative to their feeders and other details. So, that's mostly on Samsung. The operations manual is basically a guide through their GUI, teling you what each on-screen button does, without telling you ANYTHING about how data is stored or interacts at all in the control software. There is a HUGE amount of interaction as data is entered, you MUST do the setup in the right sequence, or you get a total mess! They don't tell you ANY of this in the manuals. Believe it or not, they actually DON'T tell you how the XY coordinate system runs in the machine! In fact, the X+ X- labels in the machine are BACKWARDS, as represented in the board. So, the machine coordinate system has +X to the right, as one might expect. BUT, in the placement scheme, +X is to the LEFT!
But, with a HUGE amount of help from the guy at Quad who did the factory training, I've figured this stuff out and now know how to do a setup. ALL of the users of QSA30 and CP30 machines say they manually teach all placements.
I figured out in less than a week how to create a file for my old Philips machine from the CAD data, but it took me more like 3 weeks to nail down all these details. If you don't have component profiles entered first, it screws up the import of data. Due to the possibility of errors, I do NOT want to teach placement and manually enter all the data (part type, feeder, nozzle, etc.) when the machine CAN do it. And, now that I have it figured out, it does do it.
My Philips machine was delivered to me out of a working shop, and having never even seen a P&P machine, I had it building boards in a week. (The first-generation Yamaha/Phiips manuals were not great, the 3rd revision was really GOOD!) With all the breakdowns and crappy repairs on this machine, it took me over 2 months to do the first trial board. (It seems like it was a LOT longer that that, with all the ups and downs! I was close to scrapping the machine a few times, mostly over software/programming issues!)
Jon