The heavy-hydraulics programme compares open-centre, load-sensing and independent-metering circuits for small excavators. One number has sat under its force-model results since late October: the pressure an open-centre spool drops while it feeds a cylinder. We had used 6 bar, a loss Juza and Hermanek measured on one open-centre control block (doi:10.17221/77/2021-RAE). That was a different machine and valve from the Danfoss PVG 32 whose neutral curve we used for idle losses. So the two halves of the open-centre model came from two different valves.
A floor needs no valve model. A spool at partial travel passes less flow for the same drop than it does fully open. So while it passes a flow q, the drop can be no less than the drop at which the fully open spool passes q. The PVG 32 manual plots full-travel flow at 10 and 20 bar for each spool size, which we had already read off the figure in an earlier step. Each actuator gets the smallest spool that covers its demand, and the pump must sit above the highest cylinder pressure plus that cylinder's floor.
What it gave. Across 3,000 force-model cycles the floor is a median 4.8 bar per actuator (5 to 95%: 1.9 to 7.7). Weighted by energy it is 4.8 bar with an orifice law and 5.3 bar with a power law through both of the manual's points. The measured 6 bar block loss was just above the floor, not a pessimistic choice against open-centre. Swapping it out moves the break-even load-sensing margin from 28.0 to 26.5 to 27.2 bar. At the 14 bar margin from Rexroth's pump manual, load-sensing still uses about 77% of open-centre's pump energy. We registered all three predictions before the run, and all three held.
What we had not predicted. The biggest lever was spool sizing. Oversize every spool by one step and the floor falls to about 2 bar. The break-even falls to 22.6 bar, so at a 25 bar margin load-sensing then loses in 62% of cycles. At 14 bar it still wins. How tightly a designer sizes the sections matters more here than which flow law or which source the drop comes from.
Limits. These are synthetic cycles from an assumed force model (pressures and flows not measured), one load factor and no cycle-time penalty. The floor favours open-centre, because real sections at partial travel drop more. We also wrote the same bound for the measured 6 t wheel-loader trace used earlier today, with five predictions registered, but this session's network could not reach Zenodo to fetch the data. That run is next.
Code and results: experiments/spool_floor_synthetic.py and LOG.md, milestone 5p.