Sorry for the late reply. I have been traveling for the last month, and get to spend this week with my family. I am catching up on my responses.
As to the issues that you highlight with the dual shaft motors, this assumes that you are going to mount the encoder on the shaft of the motor. There are many other places that it can be mounted. Some of these places are at the other end of a lead screw, on the opposite side of a pulley, or on another pulley that is not being driven by the motor. Each of these locations have their advantages and disadvantages.
The issue that you bring up of speed, that is a potential issue, but most likely, we will not be running in those speeds as we will be limited by other things such as the PV ratio of a lead nut, critical shaft speed of a lead screw, belt stretch, etc. It is something, though to take into account if we get there.
As to the magnets being placed in a hobby motor in incorrect locations, this could be an issue, and one that we will have to look into, but if we do a back of the envelope calc and assume that we have a 5° alignment error, and a pitch on a lead screw with a lead of .05", this would lead to a position error of ±.0007". This would still be an acceptable placement error. I will let Bob aka salbayeng comment on my math ;)
Of course, if you really need acceleration, speed, and precision, the way to go is linear motor, but they aren't cheap.
Another approach is to use steppers with advanced drives, which can extend the torque curve or detect stalling. For example, Trinamic sells chips and drives which detect stalling, and stepper drives from Copley, Nanotec, and Quicksilver can all drive steppers using servo drive techniques (and require position feedback, and cost quite a bit, although Nanotec sells integrated NEMA24 stepper + driver + controller for <$250).
QB02300 / QS02300 motor is >$500 each, without encoder (I recently looked into them, because I was speccing out a nice NEMA23 servo with 80,000 count/rev encoder).
Typical modular or kit encoders run >$60, although CUI's AMT capactive encoders are ~$25 (and require a dual shaft motor). BTW, most encoders have pretty low maximum speeds (e.g. 15000 RPM for AMT, 8000 for most Accucoders).
If you want cheap and good repeatability, you may want to think about software corrections (map the particular motor and encoder). Hand-gluing a magnet on a carelessly built hobby motor looks like a recipe for cheap, but not precise and repeatable.
BTW, I have a lot of respect for our ME's, since they build fixtures that can measure a part with a repeatability of 2 microns -- and that's removing the part and putting it back on, again (if the part stays clamped, repeatability is more like 0.2 microns).
Sorry for the late reply. I was able to travel back home, and I have been spending the weekend with the family. This is really great information for the reason that it really opens up the design space of the motor selection. I think that this combined with the hall effect, non-contact encoder from AMS that operates at high speed could make a great combination.
I was able to do some basic researching over the weekend about using bldc in closed loop control for positioning. It does seem like a promising thing to continue to look into as it could open up the speed profile for what we are trying to do.
As you have mentioned before, though there are a few other reasons to continue to look at stepper motors. Some of those include the large installed user base that already has experience with this type of technology. At the same time, though, if you continue to do what other are doing, you will result in a machine that does just that, the same as everyone else is doing. This could be a differentiating factor.
I have a few days off this week. I am going to see if I can get a meeting with the guy at TI to talk through some of the things of their solution, and what it would take to implement in this type of design. They do have a nice dev kit that I might see if they will allow us to use for a while as we really try and evaluate this option. This would allow us to really dig a little more into this and see if it is an option.
OK three types of motors used (found in my shed) , experience suggests they are of right size to do the PnP job. All ~ 60mm size (NEMA23), The servomotors are typicaly of what pops up on ebay regularly.
Yaskawa : old fashioined , nice sinusoidal windings, needed an inefficient driver as you need 200v DC to make a 140V sinewave, very low distortion on back emf, zero cogging. about 300% torque overdrive available
Allied: newer style BLDC type, a bit coggy, moderate flux distortion, designed for trapezoidal drive, has a lot of reserve overdrive capacity with 1000% torque and 600% on speed.
no-name outrunner , very little data on the web, these numbers mostly from my measurements, very non-sinusoidal back-emf (3-5%) and a lot of cogging, a lot of asymettry as the motor is spun over a full revolution (the magnets are glued on haphazardly, and the airgap is non-uniform)
The motors have different nominal voltages (200,24,10) and different nominal currents (1,9,50) although all deliver nominally 3000rpm and 0.4Nm torque, so the electrical differences are just that some have more turns of thinner wire (if they were transformers they would have different turns ratios). These differences are reflected in the Kt and Kv values (which are analogous to turns ratios).
An important consideration for a robot is the dynamic stiffness, this determines how soft the drive is at high frequencies (i.e. above the servo bandwidth). This also determines how well the position is held, when holding at a position (most servo controllers will drop to half current, and hold the last flux vector (i.e. this is open loop). So I've shown this as for example Ks=4.69oz.in/deg This actually looks better than the real servo motors (mostly because the outrunner has twice as many magnets), but it should be considered in terms of the cogging torque of +/- 4oz.in , so the actual position error of the outrunner motor, due to some external disturbance torque of say 4oz.in could be around 1 degree plus or minus another degree (due to cogging) so position error is somewhere between zero and 2 degrees.
While holding @ 50% Inom , all motors will dissipate 1/4 of "nominal" I2R losses, around 5W for the traditional servo motors, and 7.5W for the outrunner, most outrunners spin rapidly, and suck a lot of air through the windings, but in a hold situation the outrunner may overheat. (Possibly could hold at 33%, but cogging torques get proportionally more significant)
From a dynamic perspective, the motors are similar, with L/R time constant of 1.6,0.94,0.65ms so you could perform servo updates at 1kHz,1kHz,2kHz and do the PWM at 10kHz,10kHz,20kHz. Because of the irregular flux patterns of the outrunner , you will need to back off the servo gain to ensure stability, but you will probably end up with a similar GBW product as the "real" servomotors.
In summary , it looks like the outrunner could produce similar positioning outcomes to servomotors, at less than half weight and 1/10 of cost.
So you just need to plug in say 78oz.in as torque and 6000rpm , with some leadscrew pitch number , and the putative mass of the head to see what trajectories are possible with the outrunner. (Note, as is typical with most servo drives, the RPM is generally limited by external mechanical effects and bearing life, not by electrical constraints)
Had time to grab two servo motors from the shed and compare with the outrunner types.
I'm looking at nominally 200W size motors in a NEMA23 frame (60mm square) , and for the outrunner I grabbed a big 55mm outer diameter unit with a Kv of 700, prior experience these three are probably in the ballpark for your PnP project.
So I spun the motors with a battery drill and recorded the waveforms, and additionally as I had no data for the outrunner, I did a torque vs angle measurement (with kitchen scales & digital protractor) (I was going to clamp it in the lathe and use a strain gage , but the tiny amount of torque would be lost in the noise with a 1kg gage).
I'm with ZeeGlen here, on hand assembly, I use 0805 parts on a 100mil pitch, with 20mil tracks. This sort of spacing means its easier to jam in extra parts should you need them, and you can cut and scratch a 20mil track to retrofit a component.
The other tip for hand/proto assembly is to make the pads longer than needed, (i.e. hanging outside the component further) this gives you a reserve solder volume, and allows hand touchup with an iron or solder wick later. You can also put the paste on as a stripe for 0.5mm pitch parts, so that the stripe is outside the actual component leads, it will then suck up whatever it needs , the excess solder forms balls OUTSIDE the part. The longer pads mean you have somewhere to solder some wirewrap to if you need to reroute pins.
Another tip is to have some alcohol based flux is a squeeze bottle with a 28g tip, then you can apply a little drop of magic when needed. (i.e. when reworking, or with the paste striping technique).
Also ALL the signal pins (e.g. on CPU's) should be routed through a via somewhere even if it's not currently in use, much easier to do mods later (or retask the pcb) and it looks better, as the mod wiring is under the PCB .
And where two adjacent pins are connected together, join them outside the chip (direct connections tend to look like shorted pins after soldering, and if you change your mind later, you can't fix it).
I agree. We might be able to have a very basic model, as basic as it has all the slides and the vac system, and it will allow for a manual pick and place, all the way up to the full blown system where you put in the components and press a button, and you end up with a finished board.
Vision might be a good feature to make optional. I super high volume machines, the manufacturing pre-engineering will ensure that all components are in a known orientation. In mid volumes, or low-volume, high mix, as they describe it, it's more appropriate to have a vision system to ensure that everything is oriented properly.
With a small home-use device like this, vision would still be a very nice feature, but I could certainly see it being optional. The user could just take a bit of extra time to make sure that the machine knows to orientation of each component.
Having things like vision, past dispensing and reflow as options will give more options in the trade-off between time and money.