Actually working on it this week. Don't have high expectations, but will publish results, if sufficiently positive, in a few days time..

Your reference to avoid “why it is not practical” is the reason I abandoned this idea: momentum just doesn’t scale well.  But this may serve as food for thought. 

About 15 years ago, I began to work on a small hump yard using several air streams coming from under the track and directed at a steep angle (down-wind, if you will) to move the cars along.

Observations:

• With too few outlets, the cars sped and slowed noticeably as they moved across the air outlets.
  • Solution: add lots more outlets with smaller air volumes
• Longer strings of cars sitting on a track would cause the last few additions to the track to sit over an air outlet.  This would cause some cars to shake.
  • Solution: Add logic to turn on the air system to only the track(s) lined for the current car(s).  While the system thus uses less air, the complexity of trackwork greatly increases the complexity of air outlets, and sometime the previously released car should still be moving when the next car starts down the hump.
• The whole system was noisy (read windy).

I thought about, but didn’t apply the solutions above due to excess variability in car rolling characteristics and a change of modeling focus that had no use for a hump yard.  It became an exercise in “can I get it to work” versus no perceived applicable benefit to me/the layout.  Being a selfish sort, I did not consider the greater benefit to mankind and science, sorry! 

I was able to try out some cars fitted with ball bearing wheelsets this afternoon. The do roll freely almost twice as far as my best rolling cars fitted with pinpoints, but the ideal target of a "kicked" box car gently rolling 20 car lengths at a steady 5 mph didn't happen.

I'll now look into some sort of power assisted rolling, but that will have to go back on the to-do list as I expect it to require a lot of high risk design work, even before testing.

.... were available from NWSL years (well actually DECADES!) ago. They were a 40' chassis with a flywheel connected with a rubber band to one truck. Originally developed by John Allen. I had a few and "flying switches" were possible. Had to be careful with the throttle, because a car going too fast into a siding would cause a realistic pile-up at the bumper end. Also 4 or 5 in a train would put a lot of inertia, even to the point of pulling out a coupler. Unfortunately they are out of production. But fun they were!.

Dear Andy, Ken,

The 2 solutions which come to mind
(in no particular order of preference)

1 - seem to recall an old product with a flywheel in a boxcar, rubber-band-connected to 1 or more axles. car motion = flywheel spins = "momentum".
(EDIT: Pip got there first,.. ;-) )

2 - With the drives available from Steam Era Models

http://home.waterfront.net.au/~sem/bbeetle.htm

NWSL

http://www.nwsl.com/uploads/cat_chap2_for_web.pdf

and Hollywood Foundry

http://www.hollywoodfoundry.com/

and a small DCC decoder, it would be entirely possible to create a freight car with "invisible" drive,
which would allow both the engineer and the brakeman to play an _active_ roll in "fly-switching" operations.

to wit:
- the loco and any "powered" cars are consisted, and roll into town under the control of the "engineer"
- the brakeman (with his own DCC throttle) metaphorically walks down the train, closes the anglecocks
(triggers a sound decoder on the freightcar?)
- climbs up the ladder next to the brakewheel (steals control of the "freight car" from the "engineer's throttle")
- and "releases the brakes" (effectively "throttling up" the boxcar using his own "brakemans throttle"...)
- the Engineer "throttles up" the loco, and both loco and "brakeman controlled boxcar" start lumbering down the track.
(NB I don't know how they uncoupled the loco and car?)

Just shy of the turnout, the loco is braked to a stop by the engineer,
(uses his DCC throttle to stop the loco)

the "boxcar" continues on (driven by it's own "momentum"/drive mechanism)

and the brakeman "brakes it to a stop" using the "handbrake" 
(his DCC throttle, controlling the boxcar's mechanism).

For those DCC systems with 2-knobs per handset, one physical human could play both "engineer" and "brakeman" in this scenario...
(Left Knob = loco, Right knob = "gravity/momentum-driven cars")

Just thinking out loud...

Happy Modelling,
Aim to Improve,
Prof Klyzlr

I was able to try out some cars fitted with ball bearing wheelsets this afternoon. The do roll freely almost twice as far as my best rolling cars fitted with pinpoints, but the ideal target of a "kicked" box car gently rolling 20 car lengths at a steady 5 mph didn't happen

I'm wondering if more weight is the answer in this case?  I've considered the ball bearing wheelsets myself and would imagine that setting a car's weight slightly above NMRA standars may increase its free-rolling potential.  None of my cars are properly weighted (yet) but I did experiment, after reading this thread, with a topgon "fitted" with a spare battery from my oscillating tool.  The results: starting on my yard lead, I was able to kick the overweight car through 8 turnouts and about 4 car-lengths down a yard track.  Of course, the battery weights about the same as an HO locomotive and is not in any way practical, but the experiment was interesting none the less.  Slightly heavier cars with ball bearing wheelsets may produce desirable results...

-John

[T]he ideal target of a "kicked" box car gently rolling 20 car lengths at a steady 5 mph didn't happen.

You're not asking for much, are you, Andy?  I'll point out that even on the prototype, getting that sort of result is difficult.  It takes a loaded car, flat (or gently downhill) track, welded rail, a fairly straight shot, good weather, and a little luck.  We normally kick cars (assuming level track) at around 10 mph.  My technique, taught by many old head engineers, is to go straight to notch 8 when the conductor asks for a kick, and stay there until he says stop.  I let him judge how fast to kick the car, and from experience it's usually around 10 mph unless the car is only going 5-10 car lengths.  Even then, 5 mph is at the slow end.

So don't beat yourself up too hard if it doesn't work.

2 answers.

Yes I was hoping extra weight would up the momentum in proportion, but the very low speed friction (stiction), went up at about half the rate.  But a  2 Lb box car has a REALLY nice smooth rolling action, even if it still slows too fast. But I was able to move a 1Lb one by blowing on it!!! (but big puffs). Can't wait to try pulling 50 like that

Yes I'm aiming high (slow?). But the operations "feel" of moving really heavy cars over track, where the loco actually has to apply real power to accelerate even a small string, is quiet amazing. Plus the weight on track issues suddenly become very prototypical too.

...... that your best bet would be a flywheel-equipped car. Could not find a pic, but recalled that John Armstrong had one also shown in a mid-50's Model railroader. If interested I will try later to make a diagram. Real easy to make.

I have the specs of the NWSL flywheel model and the various hump yard modeling air blower and rail vibrator tricks to simulate the free cars' momentum.

But I'm a stickler for working through the validity of the many various assumptions that most of make before doing any design work. I just wanted to firmly establish the passive performance limits before the next steps, or considering adding any artificial aids. And the simpler the eventual solution the better (And usually least expensive).