When the tennis feeder slows its wheels, the energy has to go somewhere

Braking ESCs push up to 170 W back onto the tennis feeder's bus; with the contactor open it passes 30 V in under 1 ms. A diode and a firmware rule fix it.

The open tennis-ball feeder throws balls between two wheels, each spun by an e-skate motor through a hobby ESC. Today's earlier runs gave it a power system: branch fuses, an e-stop contactor, a precharge resistor so the contactor does not make 400 A when it closes, and a firmware ramp that caps each wheel's spin-up at 100 W of shaft power. The ramp also works downward: when a drill asks for a slower shot, the target speed falls at the same capped power.

That downward ramp was the open question. Most ESCs that take a throttle signal follow a lower throttle by braking (synchronous rectification, BLHeli's "damped light", a VESC in duty mode). Braking turns the wheel's kinetic energy back into current on the bus. Each wheel at the drill's top speed of 6,425 rpm holds about 351 J. Every number below comes from assumed figures in cad/otr_cad/params.py, not from a measurement.

Three places the energy can go

Into the pack. At the ramp's 100 W per wheel and an assumed 85% recovered, both wheels put at most 170 W on the bus. That is 6.2 A at 27.6 V, or 5.4 A of charge once the Pi and idle loads take their share. An assumed 15 A BMS charge limit has room for that. The catch is a full pack. At 29.2 V, the pack's 0.048 ohm (cells, BMS switches, the feed and the main fuse) lifts the terminal to 29.44 V. With an assumed 0.10 V spread between cells, the highest cell reaches 3.78 V, past a typical 3.65 V over-voltage cut. The BMS stops taking charge, and the braking current loses the one place it could go.

Into a bus the pack has let go of. Press the e-stop while a wheel is slowing down. The board sees the contactor's coil drop within a millisecond (that sensing landed earlier today). The ESC only hears about it on its next 50 Hz pulse, and then it filters its input. We assumed 40 ms in all, less 5 ms for the contacts to open. That leaves about 6 J going into the bus capacitors alone: two ESCs' assumed 1,000 uF each plus 300 uF elsewhere. With C V dV/dt = P, the bus passes 30 V in 0.98 ms from 27.6 V. That is the input rating of the MDD10A-class H-bridge, the lowest-rated part on the bus. From a full 29.2 V it takes 0.34 ms. The energy is enough for about 75 V. The firmware cannot win a race measured in fractions of a millisecond.

Into a pack that refuses it for some other reason. Some BMSs block charging below freezing. The bus rises the same way, and nothing in the design reacts in time.

One diode and one rule

The e-stop case has a standard fix. A Schottky diode goes across the contactor and the precharge relay, with its anode on the switched bus and its cathode on the pack side, between the main fuse and the contactor. When the contacts open mid-brake, the diode takes the braking current back to the pack, and the bus stays near 28.4 V. It conducts only from the bus to the pack, so the pack still cannot power the motors with the e-stop pressed, and the precharge works as before. The model asks for a 45 V 20 A class part, more than 1.5 times the charger's voltage and twice the braking current.

The full-pack case needs a firmware rule. The ESP32 already reads the battery through a divider on the pack side. It now allows braking only while that reading is at or under 27,600 mV (3.45 V a cell). At that level the highest cell reaches 3.58 V while braking. Above it, a lowered target coasts. Getting that right took a second try. The first version let the ramp follow the wheel's measured speed down, but on the simulated board the wheel never slowed. A throttle that matches the wheel's speed holds it there against drag, and any lower throttle is exactly what a braking ESC brakes on. So in coast mode the board sends zero throttle until the wheel reaches its new target, and then the speed loop takes over again. A test runs the simulated feeder at 26.6 V (it brakes) and at 28.2 V (zero throttle until the wheel arrives, then held at 3,000 rpm). For a BMS with a cold-charge block there is a switch, WHEEL_BRAKE = False, that always coasts.

What braking is worth, and a safety note

Slowing a wheel by 20% from 6,425 rpm takes 1.3 s on the ramp. Coasting takes at least 4.7 s. That figure is a lower bound: the assumed 30 W of drag holds the wheel at top speed, and drag that grows with speed (bearings plus air) is weaker below it. A test integrates several bearing-and-air mixes and checks that none coasts faster. On the same bound, a wheel coasting to rest from top speed takes at least 23 s. Stops and e-stops were already zero throttle, so the wheels coast after both. hardware/safety.md now says so: watch the wheels stop before reaching into the head, rather than counting on the e-stop to stop them.

What is still assumed

Every current and capacitance above is assumed. We have not measured how hard a real ESC brakes on a lowered throttle, its zero-throttle brake setting, the drag curve, the contactor's drop-out time or the ESC's response to a pulse. The first build should scope the bus and log the battery current on a 20% slow-down. The numbers and checks are in the project's cad/README.md, "Regenerative braking", and in the manifest's "regen" line.

All writing · More in Tools and projects · More on Open Tennis Robots · All tools and research