The heavy-hydraulics programme asks which hydraulic circuit a small shop should build into a small excavator to burn the least fuel. Its paper ends on a rule about the fixed gear pump the builder would fit. In half of 384 measured cylinder-geometry readings, a smaller fixed pump on an open-centre valve beats load-sensing once that pump's overall efficiency at 150 bar and 85 L/min passes 0.821. Earlier today a second number was added: each kilowatt of no-load drag at 85 L/min moves that threshold by about 0.019. As a formula, E50 = 0.8277 + 0.0187 (kW − 3.57).
Both numbers came from the same four pump families: three makers' catalogues (Casappa, Danfoss, Marzocchi) and one bench-tested unit. The pooled fit, the choice of the 150 bar index and the curve form were all made while looking at those four. So the honest test is a family nobody had looked at.
The fifth family
Bosch Rexroth's data sheet for its AZPG cast-iron gear pumps (RE 10093/2019-09-27) draws flow at 20 bar and at the maximum intermittent pressure, plus drive power at 50 to 280 bar, against speed, for ten sizes from 22.5 to 63 cm³/r. The charts are vector art, so they can be read exactly: 1,684 points, with axes fixed by the charts' own grid lines (largest residual 0.12 kW). Six predictions went into the log before the PDF was digitised.
What came out:
| AZPG | the four earlier families | |
|---|---|---|
| overall efficiency at 150 bar, 85 L/min | 0.912 | 0.759 to 0.913 |
| no-load drag at 85 L/min | 2.58 kW | 2.85 to 3.64 kW |
| 50% crossing (own shape, own drag) | 0.8040 | 0.8116 to 0.8297 |
The AZPG is the most efficient catalogue yet, as good as the one bench unit. It also drags the least in neutral.
The test
| rule (as committed, not refitted) | predicted crossing | error |
|---|---|---|
| one number: 0.8212 whatever the pump | 0.8212 | +0.0172 |
| two numbers: 0.8277 + 0.0187 (2.58 − 3.57) | 0.8091 | +0.0052 |
The one-number rule had claimed the crossing moves by no more than 0.015 with a pump's curve shape or drag. That held for the four families it was built on. It does not hold for the fifth. The two-number rule lands within 0.005, about its in-sample error.
Swapping halves shows why. With the Casappa catalogue's drag, the AZPG's efficiency map crosses at 0.8207, 0.0005 from the pooled number. Put the AZPG's drag on the Casappa map instead and the crossing falls to 0.8108. The cause is the drag, not the shape of the map.
What it means for a builder
Read two numbers off a gear pump's data sheet: its efficiency at 150 bar and its power at zero pressure. The AZPG at its catalogue efficiency makes the smaller fixed pump win in all 384 geometries. Other makers' catalogues read lower than their own bench tests (Casappa's by 0.15), so the paper still asks for a test of the pump, not just its data sheet.
Two predictions failed. The AZPG's efficiency (0.912) sits above the catalogue range we expected, and the one-number rule missed by more than 0.015. Four held, one of them by only 0.08 kW (the drag prediction). The limits are the usual ones. This is one more family of typical curves at 50 C, not a test report. The copy of the data sheet came from a distributor mirror, with its sha256 recorded. And the two-number rule's intercept is still Casappa's, so this is one confirming point, not a refit.
Code: digitise_rexroth_azpg.py, gear_pump_rexroth.py, results in gear_pump_rexroth_results.txt. The paper's manuscript now says this in sections 3.5 and 4.