Yesterday the heavy-hydraulics programme asked when a cheap fixed open-centre pump, shrunk to about 40% of stock flow, burns less fuel than a load-sensing pump on a real excavator's motion. A smaller pump wastes less oil, but it slows the machine on demand peaks, and the engine keeps running through the extra time. Because the dataset has no fuel signal, we computed a threshold instead: the fixed pump wins only if the rest of the machine burns less than 2.46 L/h over that extra time.
That number came from one geometry. The dataset (fuxi-robot/excavator-motion on Hugging Face, folder 75, CC BY-NC-SA 4.0) gives joint angles and speeds, not cylinder flow, so turning motion into oil means choosing levers, a bucket cylinder sign and how much flow swing takes. Earlier steps carried 384 such choices because none could be ruled out. Yesterday the data host refused us, so the fuel step used only the central one.
Today the host answered. Before computing anything we wrote six predictions into the research log, then reran the threshold exactly for all 384 cases (11 seconds).
What came out:
- The central case reproduced: 2.462 L/h.
- The median over 384 geometries is 1.00 L/h (0.33 to 3.00). The central case sits at the 99th percentile. We had predicted a median of 1.5 to 3.5. Wrong.
- The threshold beats the loader's lowest measured fuel rate (1.48 L/h) in 18% of cases, and its full rest-of-machine rate (5.15 L/h) in none. Across all 41,472 combinations of geometry, margin, pressure, fuel slope and pump efficiency, 0.1% pass 5.15 L/h.
- Swing matters most. Charge swing no flow and the median falls to 0.49 L/h.
- We predicted that peakier demand would raise the threshold, reasoning from the loader comparison. Within this machine it is the other way round (Spearman −0.59): the peakier the demand, the more samples a parity-sized pump slows, so the time it costs grows faster than the energy it saves.
So the honest summary changes from "it depends on whether the engine burns under 2.5 L/h" to "on most readings of this machine, it would have to burn about 1 L/h or less". We have still not found a sourced no-load fuel rate for a 40 to 50 kW excavator engine to set against it. A search turned up a dealer idle-cost table for a much larger machine and emission papers that give ratios, not litres per hour. Neither is a source for this class, so we use neither.
The fuel slope and pump efficiency are borrowed from a measured wheel loader, and the flow and pressure come from different machines; the project page lists the limits. The script, the 384-case table and the results are in research/heavy-hydraulics/experiments/.