Received our Neoden4 2 weeks ago (in San Antonio TX). Thanks to all the people on this thread that helped make that decision.
So far we have made 2 designs (about 8 boards total). Keep in mind this is our first go at P&P besides my 10 years or so of experience with previous company who had 3 multi-M$ lines (but I did not run them of course).
Things exceeded my expectations. Here's comments I can think of and really anything negative that comes to mind (everything else is good).
Cons:
1. We were stupid and only bought enough for 10 of each part like we might normally do when hand-placing. Should have at least spent $20-30 for the cheap resistor and capacitor reels. This led to lots of issues with feeders and hassle dealing with the shortcomings of software. Also trying to tape extenders on the peel so that it would reach the peeler (led to even more problems and less parts).
So not completely the machine's fault, but still a pain b/c software doesn't deal well. When a part fails to pick (after 3 retries), there is no "skip" part option. Only option is to try again or stop the build. If this was my software I could fix this in about 15 minutes! So you have to stop the build, and then can set where you will restart, but that's a pain if you're missing multiple parts as we were. Note that you can skip parts, but when you just want to place till they run out, the s/w sucks. So they should add a "skip" option to this prompt, or even better also add a "align" so you can feed a new part in and keep going.
Further we have found that the feeder tape needs to be fed about 2 or 3 inches so the head of the tape starts wrapping down below the machine or else it might snag for a few mm's and then have a mis-pick (and then come the woes I described above).
The "Vision Align" option in the "Mount"/run page is nice! It shows up after you align fiducials and you can hop to each part to visually inspect placement.
Note that to resume from a current part is also very weird. You have top stop the build, select your part, and then click "Save". (Save???). Anyway then it resumes from that part after it realigns. It didn't seem to always work if I tried to do this after aligning so I just got in the habbit of stopping, aligning so I could use the "Vision Align" to find the last unstuffed part, then stop, then click the part and do the save, then run again. Sigh. After a few times of this I realized that when a pick fails and you have to click "Stop" (b/c no "skip"), it will tell you which part is next in the little output window. That helped.
In regards to all this, my advice is that you don't enable the multi-nozzle pick until you are ready to run in high volume and know the feeders and parts are all good to go and not going to run out after a few boards. Reason here is because if you try 3 parts at once (i.e. Line #22, 23, and 24,), then if 22 fails to pick, you can't easily resume. You can abort, then restart on #22, but now you will replace 23 and 24. Note that as a result we had a few 0402s so perfectly placed on top of each other you couldn't tell from the top camera during "Vision Align".
Also, some parts hopped out or flipped sideways. Our machine wobbles b/c we are running it 100% speed (we will sand-bag it later), and the short 2" tapes and hacked up peel might also contribute. Tip: try to pick up very early in the hole to help this out. Wish you could slow the feeders down b/c that might also be the cause (which would suck b/c might be hard to fix).
So in summary, buy a reel or at least 10 spares of everything. Also, run with one nozzle until you build up confidence in the feeders and can run a # of boards in "continuous" mode without babysitting (we haven't really gotten there quite yet b/c we really only bought for 10 of each design). Also we are going to buy some tape extenders... we were using think kapton (sp?) tape and wrapping present ribbon to extend it but those fell off sometimes also adding to problems.
1.5. We spent a lot of time understanding how the fiducials and board alignment worked (as many people mentioned). Largely b/c we wanted to understand it perfectly. I think the real source of the confusion is that the mark setup (bottom left corner of screen) has a visual align option when in "Manual" mode, but it will set machine absolute coordinates when it really wants PCB coordinates. So just manually pull your fiducials coords from your XY data and it will be happier. Then align the panel to your 1st component in the panel setup section (this just helps it get the fiducial somewhere in its field of view during the actual fiducial/board alignment, so doesn't need to be perfect). I made a program to make this easier that will put FID1 first and will auto-load all the fiducial coords. So all you have to do is then go do the panel align to pick FID1 (which is now your first part, instead of picking "the first part in your XY").
2. Our 2 designs had various tightly spaced 0402's, and a 0.4mm QFN, and a 0.5mm QFP as the tightest components. Placement seemed good in all cases, but our new ventures in stencil ordering were the problem. Too thick and didn't window heat pads, so this led to a lot of bridges to be touched up manually. Not the machines problem of course. But my 2 weeks experience would say that the stencil/paste is a very critical aspect. We bought the higher-end manual paste machine from Neoden (about $300-400 IIRC, I don't think its their brand though), but ended up not even doing the framed stencils for these boards. Can anybody recommend a good stencil supplier for the full framed stencils?. We aligned carefully using similar thickness pcbs to sort of "frame" the PCB being pasted b/c our stencil didn't come with a frame. So ultimately our stencil machine was just a nice table-top for this.
3. Feeder count is low. We were thinking "no problem b/c if it works well and we start doing more, its cheap enough to just buy another machine". Problem is that in the meantime you will need to a lot of swaps and it takes a few feeds for things to settle (see #1) and software doesn't really make this job much easier. Would be nice if there was a notion of a part or footprint library that saved settings such as rotation, feeder spacing, pick height, place height, speed, etc. But there's not so you will have to set this up everytime you load a part. I'm sure for the veterans this isn't a big deal to get right the first time, but we spent a lot of time trying to make sense of which direction is clockwise, height, etc and so inevitably the first run or two will be trials where you are hopping back and forth between the "Mount" and the "Feeders" pages, wasting parts on each iteration. Probably not physically possible, but it would be nice to be able to go backwards on the feeders or to be able to pick and put it right back down.
4. I could not figure out the format to load XY data. For some reason everything loaded but the X and the Y. Couldn't find a sample either and didn't bother asking. Instead found their "mount" format which is easy to understand (CSV). It stores feeders, mark settings, and XY data. I made a C# program to convert my XY data and prepare feeders. This makes the alignment problems as described in #1.5 above a non-issue. Also, I added some very basic smarts so we could create a sort of part library that will be auto-loaded into the feeder settings. Also it attempts to preserve or re-use existing re-used components which would sort of be manual with the Neoden software (I suppose you could copy/paste a previous job). My envisioned work-flow was that there would only ever be one Mount job that represents the current settings in the machine. Then to load a new board you save that (thus capturing all current loader feeder/part info), import it in my program, and then load a new XY. It then tells you all new parts to load and assigns them to empty slots first (nothing fancy) so as to try to keep all feeders maxed out at all times (to minimize swaps). We'll see how this goes b/c my first time using it was really a trial and error where I added features as I found issues. But I'm already thinking we will max out our machine now just to save a bit of extra setup time (we plan this for a lot of quick protos initially). I'll try to put this in github for others at some point if someone asks. I'll give a disclaimer that it is a 1 source file total C# hack job.
5. We blew a fuse about 1 week in. Very scary at first b/c I had just crashed their software (with an invalid file format) before needing to power cycle, when it didn't wake back up. They had a spare fuse (where the AC cord comes in), and this immediately blew too. Fortunately I think someone in this forum mentioned this issue which helped calm me down. They ship with a 3A fuse meant for 220V countries, so it needs to be twice that for 110V. We got a 10A fuse and it works (but have some appropriate 5A fuses coming from Amazon).
So in summary, the machine exceeded my expectations. We did the 0201 test boards with all placed accurately but 4 on side (out of about 20) that was presumably same feeder issues from #1 where parts shake out or sideways due to machine running fast and we have a wobble.
I'll give another update once we do more than a handful of protos and actually get appropriate reels.
Did anything more come of the openpnp.org discussions? Seems this thread has died but couldn't find anything anywhere else either. I have a background in CAN (developed
http://www.cancapture.com) and reversing CAN protocols so can help with the feeders and provide hardware. I can also provide the RS232 serial logs if that hasn't happened. I would much rather be fixing problems in an open source software suite than trying to write custom C# code to work around neoden's short-comings. So Jason, if you're listening, you have a tester here. I'm afraid the learning curve would be too high for me to port the entire machine, but if you can get something started and partially running, I can definitely help debug and finish off the code. If we could get openpnp running, from what I have seen this machine could be incredible.