The Case of the Intermittent Miss-Step” of the JGB37-520 Encoder Motor

“The Case of the Intermittent Miss-Step” of the JGB37-520 Encoder Motor
—A 72-hour root-cause drill-down

Author: University RoboMaster ECU Team
Date: July 14, 2025

1. Symptom
   20 Dec 2024, during tuning of our 2025 RoboMaster infantry robot:
   • At 1.5 m/s straight-line sprint the robot drifted 3–5 cm right every 3–5 m.
   • When drifting, the right JGB37-520 (12 V, 30:1, AB encoder) speed dropped 15 %; left wheel normal.
   • Restarting the STM32F407 master masked the issue for ~30 s, then it returned.
   • Motor temperature 35 °C, supply 12.1 V steady, no over-current or under-voltage alarms.
2. Initial Checks
   | Step | Action | Result | Conclusion |
   |---|---|---|---|
   | ① | Swap left/right motors | Drift direction follows motor | Motor-specific issue |
   | ② | Scope A/B waveforms | Clean 0–3.3 V square waves | Encoder hardware OK |
   | ③ | Run open-loop (encoder unplugged) | No drift | Issue in closed-loop path |
   | ④ | Logic-analyze PID output | PID saturates at 100 % instantly | False “target not reached” |
3. Root-Cause Analysis
4. Ground-bounce in shared wiring
   Motor power and encoder GND share a 30 cm ribbon cable (motor AWG20, encoder AWG28). At >1 A current, a 60 mV ground spike appears.
5. MCU filter too narrow
   STM32 TIM input filter set to 0.1 µs → spike edges counted as valid → 2–3 extra pulses → PID reduces PWM → real stall.
6. Trigger condition
   Occurs only when PWM >80 % and motor current >1.2 A; low-speed tests never reproduced it.
7. Fixes
8. Rewire
   • Separate 24 AWG twisted pair for encoder 5 V & GND, physically routed away from power.
   • 100 µF + 0.1 µF decoupling at motor terminal to kill ground bounce.
9. MCU parameters
   • TIM filter widened to 1 µs.
   • Enable 3-clock digital deglitch.
10. Software safeguard
    • Every 1 ms verify “pulse increment × gear ratio” speed; deviation >5 % → flag “encoder fault” → revert to open-loop for 50 ms then resync.
11. Verification
    • 100 m round-trip at 2 m/s under full load: zero drift.
    • 48-hour burn-in (random 80–100 % PWM): no miscounts.
    • Seven competition matches, 3.6 km total: no sprint drift.
12. Lessons Learned
13. “Encoder looks good” ≠ “signal chain is good”; ground bounce is the silent killer.
14. At PWM >80 % the JGB37-520 can draw 1.5 A; treat its ground wiring like a power line.
15. Always add software escape: closed-loop systems need “sensor lost → open-loop safe” logic.

The ECU team turned this drill-down into “JGB-Series Wiring Checklist V2.1”, now mandatory for every future bot.