A four-layer BLDC motor driver board that worked fine on the bench started throwing intermittent overcurrent faults once it went into an enclosure near a switching supply. Classic symptom of a ground problem, but the board had a solid pour on layer 2 — so “classic” wasn’t good enough this time.
Built a small debugging rig instead of guessing: a current probe on the ground stitching vias near the gate-driver section, a differential probe across the shunt, and a slow logging run while flexing the enclosure by hand to reproduce the fault on demand. Turned out a 40MHz clock trace for an unrelated housekeeping MCU was routed straight across a split in the ground plane — the split existed to isolate an analog reference, but nobody checked what crossed it.
Fix: rerouted the clock trace around the split, and added a stitching capacitor directly under the one trace that still had to cross it. Overcurrent faults gone after 40 hours of soak testing. Full debug log, board photos, and the rig’s own schematic are in the repo.
Related: the forum thread where this got debugged in public, replies included.