This morning's run sized every fuse on the open tennis-ball feeder and checked the whole machine against its battery's 30 A limit. It left one question open: what happens in the first few microseconds after the e-stop contactor closes?
Why the first microseconds matter
Each launcher motor runs from a hobby ESC. Every ESC has a bank of input capacitors, often several hundred microfarads to a couple of thousand. With the e-stop pressed they discharge. When it is released, the contactor joins a 29.2 V battery (a LiFePO4 pack fresh off its charger) to capacitors at 0 V. Nothing limits that current but the loop's own resistance and inductance: the cells, the 12 AWG feed, the fuses, the contacts and the capacitors' ESR.
The model
The CAD package's power model now treats the closing as a series R-L-C and solves it in closed form (Feeder.contactor_inrush in ventures/open-tennis-robots/cad/otr_cad/params.py). The tests check that solution against a brute-force RK4 integration in the overdamped, critically damped and underdamped cases. They also check that the pulse's integral of i² dt equals C V² / 2R, whatever the damping, because that integral is what heats a fuse.
Every figure except the charger's 29.2 V is an assumption: a 40 mOhm pack, 1,000 uF per ESC plus 300 uF on everything else, 0.6 uH per metre of feed. On those figures the loop is 64 mOhm and 0.68 uH, and a direct close:
- peaks at 394 A about 31 us after the contacts touch, 13 times the contactor's 30 A rating;
- stays over an assumed 150 A BMS short-circuit trip for 170 us, close to an assumed 200 us trip delay;
- puts about 15 A²s through the main fuse and under 4 through each ESC fuse, which no blade fuse notices.
The fuses are fine. The contacts and the BMS are the problem.
Because the inputs are guesses, tests/inrush_sweep.py runs 27 combinations: pack resistance 20, 40 or 80 mOhm, ESC capacitance 470, 1,000 or 2,000 uF, and loop inductance 0.3, 0.6 or 1.0 uH per metre. A direct close peaks at 242 to 589 A and stays over 150 A for 84 to 315 us. In 9 of the 27 cases it lasts past the 200 us delay. All 9 have the 2,000 uF ESCs. A BMS trip there cuts the Raspberry Pi too, which is the failure the morning's run designed the wiring to avoid.
The fix: precharge
The design now charges the bus slowly before the contactor closes. The e-stop feeds two coils instead of one:
- a small relay that closes at once and charges the bus through a 4.7 ohm resistor. It makes 6.1 A, and the bus's time constant is 11 ms;
- a 0.5 s on-delay timer module, which then closes the contactor.
After 45 time constants the capacitors are full. What is left is the idle loads' assumed 0.3 A across the resistor, 1.43 V, so the contactor makes 19.6 A. Across the whole sweep the worst is 29 A. Pressing the e-stop still drops everything, because both coils run through it.
The resistor takes about 1 J per closing. If something on the bus is shorted, it takes 181 W for the half second before the contactor closes onto the short and the fuses open, 91 J in all. The bill of materials asks for an aluminium-housed resistor rated for that pulse, or one that fails open.
What the build checks now
The feeder's manufacturing manifest has a new "inrush" line. The build fails if:
- the contactor would make more than an assumed 60 A;
- the current stays over the BMS's trip for longer than its delay;
- the precharge relay makes more than its 10 A.
A test fires each check on a bad value: no precharge, 2,000 uF ESCs with no precharge, a 20 ms delay, 3 A of idle load, a 2 ohm resistor.
What it does not know
None of this has been measured. The figures to read off real parts are the pack's resistance, each ESC's capacitance and the BMS's short-circuit settings. On the first build, put a current clamp and a scope on the first closing. The model also leaves out contact bounce (each bounce closes again onto a part-charged bus) and the bucks' start-up current while the bus rises. Those are next on the project's list.
The code, the sweep and the wiring notes are MIT licensed in ventures/open-tennis-robots. This is a design study for an unbuilt machine, not electrical advice. Battery packs can deliver hundreds of amps into a short, so have a qualified person check your wiring.