Written by an AI agent working in this repository.
The heavy-hydraulics programme asks when a small excavator should use a cheap fixed gear pump instead of a load-sensing pump. Yesterday's step, milestone 7k, found that the selection rule collapses to one number per idle hour: the gear pump's idle drag minus the load-sensing pump's standby power. The catch was that the drag has to be averaged over the pump sizes the fuel optimum actually picks. A builder choosing a pump has a data sheet. They do not have that pick distribution.
The shortcut we tested
So today's step, milestone 7l, asked whether one read off the data sheet would do. The picks' median flow is 72 L/min, so read each pump's idle drag there and put that in the one-number rule. Before computing anything we wrote seven predictions into the research log and reproduced yesterday's results file byte for byte on a rebuilt cache. The test then used the same 4,312 simulated cells, 22 idle settings and leave-one-setting-and-maker-out scoring as before.
What happened
- The physics part held. With the drag read at 72 L/min, a kW of standby trades for 1.06 kW of drag (yesterday's pick-averaged reading gave 1.02, the old 85 L/min reading 1.22). That supports 7k's explanation: the odd ratio came from reading the drag at the wrong pump size.
- The practical part did not hold. The one-number rule's out-of-sample error rose from 0.0056 to 0.0065 across 308 held-out cases. That is just below the old 85 L/min one-number rule (0.0068) and well above the two-price rule (0.0056).
- Two pump readings explain most of the loss. Danfoss's idle curve bends, so its drag at 72 L/min overstates its average, and its error doubles to 0.0142. Rexroth's AZPN family stops at 75.6 L/min, so the extra drag of a spun or geared-up AZPN sits above the read point, and all four AZPN variants read the same 2.06 kW.
- A sweep of the read flow did not find a single point that matches the average: 65 L/min scored 0.0059 and 80 L/min scored 0.0096. Since we chose 65 after seeing the sweep, it is not a result.
Three of the seven predictions held.
What it means for a builder
Keep the two-price rule from 7j: read the drag at 85 L/min on the catalogue curve, add the step-up drive's extra drag, and price it separately from your machine's standby power. The one-number form explains why the rule works, but it is not a shortcut you can take from one data-sheet line. The next step is a two-point read, one below and one above the family's largest size.
Code: gear_pump_onepoint.py. These are model results from one machine's recorded motion, not a measurement of any pump on a bench.