This post is a follow up to the “pick and place” category of articles, if I can call it that, I am trying to keep my promise of discussing some commercial pick and place machines. I researched for a good entry level (non DIY/kit) pick and place machine that could cover assembly needs without a lot of baby sitting. I am not talking about high end machines here, but something like what Adafruit or Sparkfun first got into, something suitable for a startup both feature/budget-wise. Even these entry level commercial machines are easily in the tens of thousands of $. You can easily spend twice what you’d pay for a brand new reliable car. I don’t know about you but when I think of spending that kind of cash I’d want to do my homework and know everything I can possibly expect from the product. This article is not meant to be exhaustive but I hope to cover most concerns and give a good idea what is important to look for, I can only wish there was a concentrated resource like this when I first started my research. It’s a long writeup but I hope those interested in the subject will appreciate the aggregation.
First let’s look at some of the things I learned along the way, and the features you may run into or want to consider. Then we’ll look at some of the available machines and discuss pros and cons.
Construction. A pick and place uses big stepper motors to move a head at high speeds. Maybe you’ve seen the article & video of the PPM Quad vintage pick and place in action. It’s super heavy welded/cast frame gives it stability and accuracy, compensating for the heavy moving head. A frame made of a million small aluminum parts held together with screws will not be as dimensionally stable and will not allow fast speeds without it walking away from the table, and may tend to get out of whack and cause calibration issues after being used for some time. The frame and its stiffness is a VERY important part of the machine.
Belt or lead screw driven machines. Most machines are belt driven, this is mainstream for the entry market. More expensive machines are lead screw driven. They can be faster but also heavier.
Facilities: Power. Does the machine work at 120V/220V/other and does that mean you need special arrangements for it? This may be a small detail but it will add up when you are trying to get the machine installed and need to worry about many other things. Installing a 220V circuit may cost extra, especially if you hire an electrician. Smaller machines that we’re looking at (those made in the US or Japan at least) should be able to provide 120V since they are not high speed (more than 3-4000 CPH). European made machines might only come in 220V.
Facilities: Air. Does the machine require an external compressor? Or does it (also) require a vacuum pump? Most do one or the other. Does the air require special treatment or special oil free operation? Does it come with it or do you need to source it somewhere else? Another thing to worry about, and an extra cost of up to around $1K or even more.
Support, training, warranty. Again, VERY important things to consider. You are not only buying a very expensive machine but also getting “married” to the company behind it. You have to ask yourself how are you going to get support and how quickly. Can you call a number and have a human answer right away or within minutes? Does the company have a good support reputation? If you buy overseas or in a different timezone, are you going to wait 24h before you get an answer back? And is that going to be in chinglish or indglish? Think hard, you will spend hours and hours talking to support for various issues and training. How about if there is a missing/wrong part or you need an extra something, are you going to wait 2-3 days for the part, or weeks? I can guarantee that will happen.
How about training? Do you need to pay extra for any phone or training at their facility? Some companies charge extra for any extra training and require you to either travel to their facility or hire them to come to your location at a few $K per event. If the company is overseas, it’s not worth visiting assuming you’re on a budget.
Any warranty? You bet you should be covered here. But this is only as good as the company reputation. You should be getting at least 1 year of all around warranty.
Software. How do you control the machine? If the software is buggy and hard to learn you will be stuck spending time learning it instead of making hardware. This is very hard to know unless you go to an open house or ask someone else who has the same machine and is transparent to tell you the pros and cons.
Asking for feedback. Find someone else who has the same machine if possible. Ask all the questions you can and get as much feedback as possible. This is crucial. If someone else hates it chances are you will too.
Importing. Thinking about importing? Think twice and hard. In the US importing fees will quickly add up. When you’re looking at say a $50K purchase from overseas, you will pay 6.7% customs fees (a few good thousand or your hard earned cash), plus an array of other brokerage, landing, port, inspection, processing and most important transportation fees which can add to another few thousand of your hard earned cash. Then transit from factory to you will take a while, a few weeks at best from the time it’s shipped. Buying domestic will speed things up a lot. A freight load will take a few days or a week across the US. See my laser importing experience to get a taste for what to expect when you import a big crate.
Lead time. One of the first things to ask is lead-time. If there’s a great high-quality built machine built by a single guy in his garage, it might take several months before he can deliver and you may not be first in line.
Features: speed or CPH (components per hour). Probably the first thing you look after: need for speed!
In reality the actual throughput depends on many factors such as the ease of setup, replacing empty reels and missed picks (machine halts for operator intervention), travel distance to certain components. The rate killers are in order of effect: dropped parts, vision alignment, ‘slow’ speed selection for parts, nozzle changes and long travel distances to some component feeders.
Worth noting that it’s nice to have a machine that can work at more than 1 speed (the max speed). That way you can learn the machine and watch it at slower speeds and understand what’s going on. Also sometimes some components need to be picked/placed at slower speeds for various reasons.
Features: positioning (open/closed loop). Something else to consider is whether the machine is open or closed loop. Open loop means the machine uses limit sensors to detect the homing position of the axis and any travel is calculated based on that. In other words, the machine does not know precisely where the head is at any given time, but this is not a huge problem if the machine is well built and very well calibrated and the movement is deterministic (no obstacles expected, etc). This is typical of laser cutters and most hobby market CNCs, and they work well. However on a high precision CNC you probably want closed loop control which uses linear encoders to know at all times where the head is. That means if something unexpected happens which stalls the head movement even by a tiny bit and offsets it from where its calculated position is, all picking and placement from that point in time is wrong. When the accuracy is a mil or fraction of it, that starts to matter.
Features: Component alignment.
This is one of the most basic and important things your machine needs to be able to do. If your machine can’t align the parts do yourself a favor and move on. Look at the above strip of components, or at the camera image on the right. They are twisted and shifted in the pockets. You might train the machine to pick from the center of the pocket but if the component is out of place, it might twist and turn on the nozzle while it’s picked or during travel. Even if it doesn’t when it’s placed it will not be centered to the drop location. You might say that reflow fixes that, in most cases it might, but it’s those times that doesn’t happen that will drive you crazy when you have a panel with a few hundred parts and you don’t have an AOI machine to check it for you.
So what alignment methods are out there?
Alignment: Vision. Everybody thinks that a machine with vision is superior. You know what? They are right, and here’s why. There are 2 kinds of vision: top (looking down) and bottom (looking up). You will almost never use the bottom vision unless you have very fine pitch or BGA components, and I believe high end machines use lots of bottom cameras to align many parts at once. Top vision is used for training the board and component feeders, and also for fiducial corrections (global panel or local board fiducials) and bad board mark recognition (when a PCB in a panel is marked as bad and should not be populated). A machine that only uses vision for component alignment is slower because it has to travel to a fixed location where the upward looking camera aligns/rotates the part. Again, high end machines don’t count here.
Alignment: laser centering.
This was what the PPM Quad machines use: a Cyberoptics laser centering device. Laser centering is also available on other pick and place machines. Don’t buy it if it’s an option. It’s expensive and not that accurate. The way it works is a laser device casts a shadow of the component on a CCD sensor which measures it as a voltage graph, from that it calculates what the position on the nozzle is. A Cyberoptics laser curtain is about $7K as an OEM part and will probably add $10K to the cost of the machine and you will not gain anything by getting it and you will loose some functionality and speed on some parts. Long before laser centering or optical centering existed pick and place machines existed and they used centering fingers and wells. The advantage is that this can be done while the head moves.
Alignment: centering wells and fingers.
Assume you have a component on the nozzle and it is held in place with vacuum. Also assume the hold is strong enough for the part to not fall off but loosely enough so the part can move if pushed firmly. If you pick a part and it is within 40 degrees of the correct position you can center it using a well with precisely machined edges. You pick the part and move it over the well and lower the nozzle so the part is below the walls of the well. You then move the center of the nozzle to half the part width from the +X wall and then half a part width from the -X wall. The part has no choice, it is centered in the x-axis. Do the same for the other axis and it is centered in x and y. This is the original way in which parts were centered.
Centering fingers work similarly but with refinement. While the machine travels to the destination the part is lowered to the height of a set of fingers fixed to either side of the nozzle. The fingers are geared from a common motor so that they close on the nozzle at the same rate from each side. Normally the fingers are spring loaded and held open with a motor and a cam. As the fingers close on the part they will center the part on the nozzle with one finger pushing the part against the other finger. The fingers are opened and the part rotated 90 degrees and the fingers closed again. The part is now centered on the nozzle to within the tolerance of the fingers, way better than 0.001″ and probably around one tenth of that. This works better than the laser centering.
Centering fingers vs linear optical centering. For laser centering the part casts a shadow on a linear optical sensor and by rotating the part they can find the minimum and maximum length of the shadow. The minimum length is the smallest dimension of the part and the maximum is related to the shadow of the diagonal of the part. From the two measurements you can calculate the position of the part on the nozzle if you assume the part is rectangular. (You also need to know the position of the shadow on the sensor). That’s the rub, not all parts are rectangular or cast a shadow that allows the position to be calculated accurately. A good example of a part that is a pain is a mini-USB connector. It has feet that stick out one side and they are not in the corners of the part. That throws off the calculation. With the fingers you change the centering height to be at a level where the fingers miss the feet of the part and it centers precisely. Another part that is a major pain is the SOT-23-3 and SC70 packages with three pins. The side with 2 pins make a better shadow than the side with the single pin and it screws up the calculation. On fast, expensive machines they have multiple up looking cameras and all parts travel over a camera and the image processing calculates the center and rotation angle of the part on each nozzle. Laser centering forces you to use the up looking camera more often than the fingers. But you won’t see this on high end machines.
Alignment: none. Yes, as we will see, there are such machines. They pick and place just like the name says. Run away.
Features: nozzle changer (or not). In all cases I’ve seen, there are multiple nozzles for different part sizes. But not in all cases the change is automatic. Some machines stop when you have to change the tool manually. Do you want to sit there by the machine waiting every 5 minutes for a tool change? Let’s just make this short and say you want this automatic. Usually the tool changer comes with tool detection and also component detection. That means that if a part is ever dropped the machine detects the change in nozzle vacuum and retries the picking.
Features: dispensing option. Some of the machine I’ve seen have this option but typically it adds up a lot to the price, up to around $10k. I use stencils to apply paste so I didn’t want this option. You can see how the Quad dispenses paste before it populates in this video:
Features: PCB holder. How does the PCB sit in the machine? It should have some way of being placed at the same spot every time. That way no matter what panel you place, one corner is always the same, and only the other side moves. Typically this is the same as a tray holder but not always. Also worth mentioning the Quad used suction cups to hold the various things on the working area, including the PCB, a nice thing but wondering if not overkill.
Features: Z-axis component height adjustments. This means the head can be adjusted to travel more or less when picking/placing/aligning depending on the component height. For most very small/short components it won’t matter, but for thicker components (think SOT23, 0805 and taller) you don’t want the head to slam into the tape when picking or into the PCB when placing. Some components sit deep in pockets so you need to be able to travel deep enough for a good pick. The nozzle is spring loaded anyway so it won’t be damaged when dropping but you might bend pins or drop parts incorrectly if the Z height of the component cannot be or is not set correctly.
Feeders. This is a very big topic to discuss. Basically there are a few options. There are machines that don’t have feeders and advance the tape using the pick head, a bit weird but ingenious considering that the next best thing is a real feeder. That also means you can’t use tubed components unless you improvise somehow. Block feeders are a nice feature for very often used components, you load/unload all of them at once and such block feeders often stay in the machine all the time (common resistors, 0.1uF caps, LEDs etc). Some machines only have block feeders for taped components, I think that’s undesirable, since almost all designs have unique components. Individual automatic feeders are most often the best option since you have the most flexibility and the feeder does everything for you, including the advancing and the cover tape pulling. I’ve seen some machines that use weights to pull back the cover tape. You will sit by the machine raising the weights all the time, not ideal. Vibratory feeders are meant for parts in tubes. I’ve noticed a lot of components are cheaper in tubes, so being able to add a vibe feeder will come in handy when some component is only available in tubes. I think tray holders are available or can be improvised on any machine. The tray the components come in aligns all the components. You then train the machine the first and last component in the tray and tell it how many there are, and it calculates everything for you.
Most machines offer cut strip holders which are expensive. If you have access to a laser cutter you can make your own from acrylic. Or you can even stick tape down on a cardboard with double sided tape and then train the machine to pick them as though from a tray (first + last + count).
Intelligent feeders can be helpful when you want to keep track of your parts inventory or let the machine issue a warning or stop when the component count reaches zero. Some machines require special arrangements or cost a few $K more for this feature while others have it built in. Since a feeder can know exactly how many parts it advances so it should be a nobrainer to calculate remaining components given a starting total.
Feeder capacity, limits, indexing. In most cases the feeders talk to the machine via some kind of connector interface. Typically they are spaced such that each allows an 8mm feeder (narrowest kind) to fit. Inserting wider feeders means that it will take 2 or more slots in the machine, cutting down on feeder “capacity”. So when they say the machine ca take N feeders, it means N 8mm feeders. Naturally you want as many feeder positions as possible. A large count here directly translates to a large machine. Standard tape sizes are 8, 12, 16, 24, 32mm, 44mm. Individual feeders are expensive. Block feeders are high priced but they cost less per individual lane when you do the math. If you’re looking at a machine that has both options, it’s a win-win and you can combine block and individual feeders. Indexing refers the the spacing between components on the tape. This is also a standard and they go all the way from 2mm apart (think 0201 components or some rare small 0603 LEDs), to 4mm (most 0603 components or things like SOT23), 8mm, 12mm, 16mm etc. Nothing very unusual here, I think any decent machine should be able to do all of these. Worth noting that some machines will index 2mm as 4mm but pick twice in between, that’s fine as long as it works.
Feature: components range. Most likely a front page advertised feature. Most machines advertise down to 0402 if not 0201. But I doubt all can do 0402 well. My smallest components are 0603 and I like to stay above that whenever possible. I would say a decent machine should at least be able to do 0402 pretty well, but it’s hard to check unless you ask someone who has done it before on the same machine. The limits of advertised features might work, but not always very well. You also got to ask yourself: can the machine feed large/tall/weird parts?
Feature: software/OS. You might hate windows but most machines operate on windows. You will love it once you see machines running DOS. You want to be able to plug in a flash drive and save your stuff and also import centroid/XYRS placement files that contain your components location data. These are usually CSV or tab delimited files that can easily be imported in the software (a nice-to-have feature).
Let’s look at a few machines that I stumbled upon in my search and discuss some pros and cons. There might be others and you’re welcome to send me more suggestions but I think these machines cover a wide range of features worth considering and should be a good pool of choices to choose from. To me size is also important because of my location so I’ll touch on this.
Neoden TM220/240A. Some people bought this, and I was very interested in it at one point. Dangerous Prototypes (photo credit) got one and also Circuits@Home and RaysHobby. Most of the videos you watch are in chinese, but I could make out that most of the times there needs to be some “hand work” afterwards. That’s when I learned about alignment and the lack of it in this machine. Also everything is based on coordinates, and a red laser cross is used to align the board. This is pretty hopeless, especially when you have a larger panel with many boards. If your panel is V-scored, the scoring might introduce very minute shifts that propagate to larger shifts across the panel. That means that without any fiducial recognition or ability to adjust each panel and board position, the placement can get increasingly inaccurate. I haven’t seen this first hand and those who do can speak to or against this theory. Then I saw this set of teardown pictures and how cheaply it’s built. I consider this a toy, and a very expensive one, not really a machine. There are many threads in forums around the web discussing this machine.
Pros: size, dual head (although this is more to compensate for lack of nozzle changer), not much else
Cons: too many to list including price ($5000 … you’re kidding me?).
MDC 7722FV. This is the famous first machine that Adafruit got. They still have a dedicated page worth checking out with information about it and other things concerning PNP machines. They now got very high end machines and donated their old one to NYC Resistor. This is a cute japanese machine that does what it’s supposed to and it’s pretty well documented. See manual here. From all I could tell it’s a decent machine.
Pros: individual feeders, desktop size, if it’s good enough for Adafruit is it good enough for you?
Cons: open loop, requires special separate compressor, frame construction, imported, vision alignment
Madell SX1010 and other models. These are imported chinese machines. The price range was attractive given the advertised features but the hanging weights that pull tape and skeletal aluminum construction made me walk away and realize the price doesn’t justify the features. I also found some discussion and feedback on this machine which put the last nail in the coffin. I know someone who has this and he gave mixed feedback but it served him well for a few years before moving in a more capable machine. To me the deterrent was the hanging weights:
PPM Quad 4C/4000C. I looked at this machine long and hard. This is a 1990s vintage machine that was mass produced and is of high quality. It cost upwards of $150K originally and now sells refurbished for around $30K with Windows7 (no feeders). I learned about it from Ohararp who bought the 4C DOS based one and wrote a nice article detailing his experience. No doubt they are a great company and will offer great support but their training comes at a cost, whether onsite or at their facility. My big problem is I have to be able to get the machine down a twisted stair case that has a 32″ door frame at the end. I got very excited when I learned that it’s possible to split this machine in 2 parts. The bottom part is the larger one but the top is heavier (from what I’m told) due to the 1″ steel base. I even went as far as making a wooden frame matching the larger bottom side to see if it would physically fit down the stairs, which it did! So at that point I decided to go and visit them and they were very nice. Only then I realized how heavy the thing really is. I was betting that a team of 4 big guys will be able to carry this down my twisted stairway, but I realize that would have been almost impossible even if the bottom would have physically fitted (albeit narrowly). The big issue after taking the top apart was recalibrating the bottom vision camera which is part of the bottom side, not something I was very comfortable doing. I was also a bit concerned that during the demo we were given, there were several rejects which seemed like a high percentage, and this was on resistors and ceramic caps which are the easiest of the easy parts to pick and align. The last thing you want is for 4-5 parts to be rejected out of 100, you will never want to be salvaging components from the reject bin and placing them back in the feeder, that’s a nightmare. Not sure why that was, maybe the laser centering was not done right or something. After the demo placement all the parts were beautifully aligned on their PCB locations. The bottom of the machine is full of analog vintage looking hardware. I think all that stuff was doing what a PC with a simple vision card can do today, I know so. Everything about it was massive and hence the big heavy tubular welded steel frame to hold everything still and allowing high speed movement. Heavy is good when you can accommodate it. It can be a great machine if you know it well and it felt like it was very high quality built (made in USA in the 90’s). The windows software package costs an extra $7500 ontop the base $16K price (self install or $10K installed). Feeders are pretty cheap compared to other places (refurbished feeders even cheaper) but you should buy as many feeders as your budget can fit. You will reach $30K in no time considering you want Windows and need the training. It may be a great option and I thought it was until I realized it would be impossible to move it where I want it and then reasseble/recalibrate it myself. Leaving such a fine machine in the garage would be suicide. Heat in the summer and cold in the winter will cause rubbers/belts to deteriorate and condensation to ruin the electronics and optics. Don’t even think about it.
Pros: exceptional quality, heavy build (pro if you don’t care about this and have the facility to unload it into), lead screw, dispenser option, many feeders, conveyor loading option, flexible work space, intelligent feeders optional (if I recall, all feeders are IQ but the machine is not so need the extra stuff in the machine to talk to IQ feeders)
Cons: heavy build (con if moving it is a problem), laser centering was interesting but kept me wondering how accurate it can be after seeing a bunch of rejects.
Autotronik BS281 (aka Manncorp MC-400). This is a desktop size machine but comes with a metal stand which is great. See the product page here. Worth to mention that this machine is used by Small Batch Assembly, a startup that does small run PCB assembly. This is a rebranded Autotronik BS-281 german made machine, so I could only expect very good quality, watching this demo video (also see this this action video of a MC384) pretty much confirms these are quality robots. I don’t have a lot of feedback on the MC-400 except judging the features. Sparkfun first got a MC384/5 I think, which are much larger heavy unibody steel frame, and sometimes they have some of these refurbished for a deep discount but it’s a gamble. From all I could gather it sounds like Manncorp is a really great company, and I can confirm that feeling from my few email conversations with them. I have no idea how good and easy the feeders are but my guess is they must be pretty decent.
Pros: brand/quality, US distributed & supported by Manncorp, optional stand means you don’t have to build a strong stand yourself
Cons: distributor pricing
Mechatronika M10V. This is another machine I’ve been very interested in. I’ve only received really great feedback from some people that own it and it sounds like it’s of great build quality, and there is good feedback on Adafruit’s SMT tech forum. The drawback is it only works with two block feeders that are configured when you order. If this machine had individual feeders it would be really great. As it so often happens, if you run out of lanes and need that 1 extra feeder and don’t have it, you need another block of feeders, and not only that but the huge inconvenience to exchange blocks in the middle of populating a board because some components are on another block. The good thing is they have bigger/faster machines (cost a bunch more too) which accept more block feeders (a max of 4+vibe?) which are compatible between all the machines. Worth to mention they also manufacture reflow ovens. When I asked about lead time it was going to be about 3-4 months, shipping was around 1200Euro via DHL or Shenker, and this is not a heavy machine. Another concern was max component height which was something like 10-12mm if I recall correctly. Some aluminum caps will be more than that, I know for a fact. Nothing major though, they might fix that.
Pros: build quality, size, vision, closed loop, full vision
Cons: takes 2 block feeders only (plus vibe also available), imported (no US distributor), lack of documentation/media, 1200-1600cph – vision alignment cuts speed in half what it could be, 230V supply only
DDM Novastar LE40V. It was one of the first machines I looked at and dismissed when I didn’t understand component alignment very well. It uses mechanical centering jaws to align components which I explained in the features section. Once I saw the Quad laser alignment procedure it seemed mechanical jaws might be more deterministic, it turns out quite so. The company is nice and responsive, feedback also confirmed the quality and support. I think this is a winner.
Pros: made in the USA, support, solid welded frame, size, closed loop, full vision, individual + block 8mm feeders (96 feeder slots), 120V, includes gast vacuum pump controlled by the machine, desktop size, DDM offers open house events twice a year, free onsite training
Cons: slower than LE60V which is unibody supporting larger motors
ESSEMTEC CSM7100V – Essemtec is a Swiss company that makes very high quality machines. There are some older models that show up on ebay from time to time but I don’t recommend anyone buy from ebay or used without certification, it’s a huge gamble. Either way I thought I’d mention it here anyway since I believe these are great machines.
DIMA MP-100. Check the flyer here. Probably a very good machine with lots of feeding options and great accuracy (if it’s real as claimed). But size-wise it was not an option. Also imported. Vision aligned with laser option.
In a way I’m surprised that the options are so limited and so expensive. I kept hoping for a commercial open source machine that doesn’t suck, we’ve seen some shy attempts (Board Forge, this recent THP entry, and some others I can’t remember). The project is hard because frame is important, then feeders, vision, and an intuitive GUI that’s not hard to learn. Nobody wants to buy a machine that needs constant fixing and adjusting and that after months of movement and vibration goes out of alignment because of poor frame construction.
I hope this is good coverage for anyone interested in getting into assembly. If I missed something or something I stated is wrong let me know. I didn’t discuss pricing because I don’t know some of them and they might differ on an individual order basis, also some companies might offer discounts the more options you bundle. Consider this article open ended and subject for change. I might add more machines as I learn about them, let me know if you have suggestions.
Low Power Lab LLC