Hi Sander, All,
I've done some experimentation and I am able to reproduce the failure on demand. I can also reproduce a remedy on demand. However, the remedy is not really practical.
Current Feeder mechanism: I'm using Bertho's rolling feeder clamp design to eliminate problems caused by changing filament diameter.
The filament is jamming inside the Bowden Tube - previously we dismissed the possibility of this type of friction, however the amount of friction generated is significant and is a strong function of the diameter of the spool used to coil the filament. The spool diameter should match the curvature of the Bowden in service to minimize points of contact between the filament and the inner wall of the Bowden.
When the spool diameter is much smaller than the curvature of the Bowden it will adopt a coarse spiral inside the tube that is pushing outwards along it's entire length. Since the spiral inside the Bowden is a significantly smaller diameter than the natural state on the spool this "normal force" can be significant. Since this normal force can be high, the sliding frictional force will also be high due to the large contact area. This is observed by a steady increase in frictional force as the filament is forced into the Bowden. With a small enough coil the filament will jam in the Bowden so tightly, that the filament cannot be moved even with extreme force.
You can check this with a simple experiment testing the potential extremes - heat PLA with a hairdryer with just enough heat to coil around a cardboard toilet roll, let it cool and try feeding the filament and get an idea of the "feeding force". It will be very high and will jam in the Bowden.
Snip off another length of filament and this time use the hairdryer to make a perfectly straight filament section. It will be very easy to slide.
Remove the filament, and add a small drop of water to the surface of the tightly coiled filament - the surface tension will show the points of contact between the filament and the inner wall. It will be pushing the inside wall along the entire length. Repeat with the straight filament - there will be random points of contact.
These observations explain happy printing at the start of a spool, but jamming as the spool is consumed and the coil diameter reduces - I have experienced this. It also explains the success of UM users who routinely "respool" their filament to create larger coils, when operating from the smaller diameter spools they experience jamming - check out google groups http://tinyurl.com/9b489ez comment by WThM de Groot. Furthermore filament that has not been wound cleanly onto the spool will have slight kinks of tighter diameter, which in turn increase the "normal force" and hence dry friction on the Bowden Id.
A partial solution (also a business opportunity for UM) is to use spools with as large a diameter and as wide as possible to ensure that the coil radius is constant for the entire length of filament. You could sell empty spools with this geometry.
Another solution is to reduce the sensitivity of the Bowden system to the coiling effect, but I have no practical suggestions on how to achieve this without eliminating the Bowden entirely.
My remedy (though impractical) is to preheat the filament to get it to "relax" so that the coiling has a larger radius. I've successfully remediated filament that I would have otherwise thrown away due to jamming. A hairdryer seems to work best, to apply the minimum amount of heat. It is possible that heating may change the filament diameter - I've measured an increase in average variation from 2 microns to 20 microns, however I can print where otherwise I would not...
I just have to respool my filament or change to a different supplier who supplies larger spools and winds the filament on without filament kinking...