Harmonic drive 4th axis build

You have
DEFAULT_ANGULAR_VELOCITY = 50.000000
So it seems that probe_basic isn't using that. It might be fixed? You would have to ask there.

The accelerations don't look low (I had already checked) but you could try pushing them a bit higher to see what happens.

Can you explain again which code you are seeing the problem with?

In an earlier post I only saw one B axis move,
N81115 G93 B-982.046 F269.4382
And that isn't a combined move, so perhaps I am on the wrong track.
Anyway, without knowing where the B axis was before the move, it's hard to work out what the speed should be.

Then for example
N81175 G0 Z21.
N81180 G1 Z15. F1100.

So a move of 6mm in 1/1100 of a minute, ie 6600 mm/min

I think that it must be due to falling back to the older single-segment lookahead.
It seems to be slower when doing combined Z and B moves and faster when Z is not moving.

Looking a the "slow" G-code it is requesting 207 degrees/sec      

delta ZZ vel (mm/sec)delta BB vel (deg/sec)Z11.104B-90.454F27344.8538    Z11.099B-90.859F30739.69230.0052.56164102-0.405207.4929Z11.094B-91.264F30702.33070.0052.55852756-0.405207.2407Z11.089B-91.618F35078.19650.0052.92318304-0.354206.9614Z11.084B-91.973F35055.96770.0052.92133064-0.355207.4145Z11.079B-92.329F35033.21990.0052.91943499-0.356207.8638Z11.074B-92.633F40846.69320.0053.4038911-0.304206.9566Z11.069B-92.938F40823.12950.0053.40192746-0.305207.5176Z11.064B-93.294F34968.8820.0052.9140735-0.356207.482Z11.059B-93.6F40770.33180.0053.39752765-0.306207.9287Z11.054B-93.855F48897.10910.0054.07475909-0.255207.8127Z11.049B-94.161F40724.67460.0053.39372288-0.306207.6958Z11.044B-94.467F40716.77050.0053.39306421-0.306207.6555Z11.039B-94.722F48805.88490.0054.06715708-0.255207.425Z11.034B-94.978F48779.42860.0054.06495238-0.256208.1256Z11.029B-95.234F48753.1790.0054.06276492-0.256208.0136Z11.024B-95.541F40603.31770.0053.38360981-0.307207.7536Z11.019B-95.746F60871.43590.0055.07261966-0.205207.9774

(Edit, this was a neatly formatted table when pasted from Excel..) 

The Z step-down is _tiny_ which is probably part of the problem. The system will only reach a speed which will allow it to stop before the end of the next segment (in case that is the last segment. This is what single-segment lookahead means) Can you persuade the CAM to make the Z steps 10x larger? That should help. 

To stop in 0.005 mm with an acceleration of 800mm/s the speed can not be higher than 2.6mm/s
To stop in 0.35 degrees with an acceleration of 4000 the angular speed can not be any higher than 50 deg/s 

(v^2 = u^2 + 2as equation)

Again, these tiny steps make no sense, the system would smoothly interpolate between steps 10 or 100 times bigger. 
 

It got lost when the editor destroyed the data table, but every Z move in the slow section was 0.005 mm, ie 5 microns.

Googling on "Fusion CAM segment length" suggests that turning on smoothing might help.

 

Try 0.1mm and then try cutting air, just to see what happens.