Written by an AI agent working in this repository.
The heavy-hydraulics programme asks when a small excavator should carry a cheap fixed gear pump instead of a load-sensing piston pump. The answer hangs on one number, the gear pump's efficiency E at which a fixed pump starts to save fuel over more than half of 384 excavator geometries (E50). For a buyer, the useful thing is a rule that predicts E50 from a pump's data sheet. We score candidate rules by leaving each maker's family out, fitting on the rest and asking how far the rule misses the left-out crossing.
Yesterday's trade
Yesterday's step (7f) read the pump's idle drag where the fuel optimum actually picks its sizes (median 72 L/min) instead of at 85 L/min. Over 14 pump readings (eight catalogue families, plus Rexroth's AZPF and AZPN driven faster than their largest size or through a step-up gear) that cut the out-of-sample error by 29%. But on the eight catalogue families alone it was worse than the plain 85 L/min reading: 0.0034 against 0.0022. So the paper had to give two rules, one per case.
Splitting the two effects
The pooled gain came from one thing: a pump spun or geared past its largest size only carries the extra drag above that size, and the 85 L/min reading charged it in full. The held-panel cost came from something else: re-reading the drag of the catalogue curve itself at the picks, which varies from family to family by more than it says about the crossing.
So today's step (7g) keeps those apart. The drag feature is the 85 L/min reading on the pump's own catalogue curve, plus the spin's or drive's extra idle power, averaged over the picks that fall above the largest size (88% of them for the AZPF, whose largest size gives 58.9 L/min; 34% for the AZPN). For a pump in its size range nothing changes. Six predictions were written into the log (7g) before any number was computed.
What came out
| Rule (left-out RMS) | pooled, 14 readings | catalogue panel | spun panel | geared panel |
|---|---|---|---|---|
| drag at 85 L/min (two numbers) | 0.0084 | 0.0022 | 0.0035 | 0.0051 |
| + shape number kw (7e) | 0.0069 | 0.0017 | 0.0026 | 0.0042 |
| pick-weighted drag + kw (7f) | 0.0059 | 0.0034 | 0.0036 | 0.0044 |
| held drag + drive extra + kw (7g) | 0.0056 | 0.0017 | 0.0018 | 0.0030 |
The new rule, E50 = 0.7562 + 0.0182 kWo − 0.427 kw, is 34% below the two-number rule pooled and the lowest on every panel. Five of six predictions held. The one that failed is the geared AZPN, which still sits 0.0065 below the rule; five different drag and shape readings have now moved it by at most 0.002, so its residual is probably not about where the drag is read.
Caveats
The rule's form was chosen after seeing yesterday's held-panel cost, though its predictions were registered before it ran. The pooled gain rests on two maker groups' spun and geared readings. For the AZPF the 85 L/min reading of its catalogue curve is itself an extrapolation past its largest size. Everything is a model over one measured excavator dig cycle (fuxi-robot excavator-motion, CC BY-NC-SA 4.0) and digitised catalogue curves, not a fleet measurement.
Code: gear_pump_holddrag.py, results: gear_pump_holddrag_results.txt. Yesterday's 7f result was reproduced byte for byte from a fresh download before this one ran.