and it appears to be impossible to ramp from zero rpm without a jerk at its minimum speed.
With a duty of ~228 out of 255 the minimum stable rpm is ~90 with 6V power.
Any slower PWM duty and it stalls, but then the driver detects stall and attempts to restart causing a jerk every ~4s or so.
Turning the PWM control HIGH to enable always causes a high speed initial jerk so no real smooth start is possble.
Startup caveats observed in bench testing:
The BLDC driver’s first valid startup command can produce a large self-start burst that is not strongly controlled by small PWM-duty changes from the DSP
Bench logs showed similar initial FG spikes across several different startup strategies and duty values, including slower-than-normal launch attempts.
Separate low-speed maps, startup ramps, duty clamps, intermittent startup gating, and threshold changes did not materially remove the first-launch jerk.
Once the motor is already spinning, the outer shaft-to-PWM mapping and FG trim can regulate it reasonably; the dominant problem is the internal driver’s first commutation/start event.
Is the 3625BL 36d-25l-brushless-dc-motor any different?
If the same driver chip then likely not…
I don’t care if it has a minimum RPM of 50-100, but the inititial power-up jerk of the FT8213 is unacceptable.
Another BLDC or stepper with 0 to 1300rpm and external controller (I have a few common stepper controllers) would be worth testing if anyone has a suggestion.
Max torque needed is tiny as it only has to spin the speedo’s internal magnet (1000rpm == 60mph) and ODO gear, so essentially fan-duty.
On the 3625BL: our product page does not list the driver IC, and the attached datasheet does not either, so we can’t confirm from the listing whether it uses the FT8213Q or a clone. You are welcome to ask E-S motor directly by clicking the “Ask the vendor” icon (third from the left below he product image).
That said, your instinct is probably right, and the answer may not matter much. What you’re describing reads as normal sensorless trapezoidal BLDC behaviour rather than a fault specific to that chip. With no rotor position information at standstill, the driver has to align the rotor and then force an open-loop ramp until back-EMF is strong enough to detect. Both phases run on the driver’s internal parameters, not your PWM duty, which is consistent with your finding that duty clamps, slower launches, and startup gating all produced similar FG spikes. Any integrated-driver BLDC in this class is likely to behave the same way, so swapping to the 3625BL is a low-probability fix.
Running a DC or BLDC motor at slow speeds is not easy at all (our Lynxmotion LSS servos use DC motors). To get controlled motion from zero, it’s likely that either the controller needs rotor position feedback, or the motor type needs to not have to search for the rotor at all. Three directions, roughly in order of least work for a load this light:
Small brushed DC motor with encoder. There is no commutation state to establish, so it follows PWM from standstill with no start event. At fan-duty torque you can close the loop with the outer PID you already have. Lowest-risk swap.
Bipolar stepper with a microstepping driver. Also starts cleanly from zero, and you already have drivers. The caveat is the top end: 1300 rpm is roughly 22 rev/s, about 4,300 full steps/s on a 1.8° motor and much more once microstepped. Achievable at a higher rail voltage with the right driver, but torque falls off steeply with speed, so bench test before committing. Our Lynmotion PRO smart servos use stepper motors which are optimal at very slow speeds.
Sensored BLDC with an external FOC driver. Add a magnetic encoder at the shaft end and drive with a FOC controller. FOC produces torque at zero speed with no open-loop start ramp, which is exactly the 0 to 1300 rpm behaviour you want. More integration work, and it needs shaft-end clearance for the sensor magnet.
One workaround for the current setup: if the motor can start once at ignition-on and stay running for the rest of the drive cycle, the first-launch burst lands during the window where many gauges do a self-test sweep anyway. It does not fix the low-speed floor, but it moves the jerk somewhere harmless.
