A gearbox is not only a mechanical part. It is also a stack of problems your control code spends its life fighting. Backlash at every reversal. Compliance you tune around. Couple the motor straight to the load and most of those problems leave, and a few new ones arrive to take their place.
A direct drive motor connects the rotor or forcer straight to the load, with no gearbox, belt, or lead screw in between. Here is what changes once that transmission is gone, from the point of view of the code that runs the axis.
Backlash stops being a software problem
Gearing gives you backlash, the lost motion you feel at every direction change. In firmware you pay for it with reversal compensation and settling time you can never fully tune out. Remove the gearbox and the gear backlash is gone at the source, so that compensation code goes with it.
Be precise about the claim. Direct drive removes the backlash the gearing introduced. It does not remove compliance elsewhere in the axis, in a coupling or in the structure, so a stiff mount still matters. What you get back is a mechanical path that does not lie to your position loop at reversal.
You inherit the full load inertia
This is the trade. With a gearbox, the load inertia reflected back to the motor is divided by the square of the gear ratio. A 10:1 stage divides it by 100. Take the gearbox out and that division is gone. The motor, and your position loop, see the full load inertia directly.
That is why direct drive motors are larger and cost more at a given torque, and why the load-to-motor inertia ratio now sets your loop bandwidth and stability margin. You are not tuning against a comfortably reduced inertia anymore. You are tuning against the real thing.
Commutation moves to the front
With no gearbox in the way, smoothness of force or torque comes down to the current loop and to knowing the rotor position within one electrical cycle. In field-oriented control, torque tracks the q-axis current. That holds for a surface-magnet PMSM. An interior-magnet motor adds a reluctance term, so treat it as a close approximation, not a law.
# Simplified. A real drive runs this in the current loop at tens of kHz.
# 1. Commutation needs the rotor angle within one electrical cycle,
# referenced to the correct alignment offset.
theta_e = pole_pairs * encoder.angle() + commutation_offset
# 2. In field-oriented control, torque tracks the q-axis current.
# True for a surface-magnet PMSM. Interior-magnet motors add a
# reluctance term, so this is an approximation, not a law.
i_q = torque_command / torque_constant
i_d = 0 # no field weakening in this example
# 3. With a gearbox you would divide load inertia by the gear ratio
# squared. Direct drive removes that division, so the position loop
# is tuned against the full load inertia.
Two feedback jobs live here, and developers mix them up constantly. One is coarse rotor position used to commutate the motor, often from hall sensors. The other is the fine axis-position feedback that tells you where the machine actually is. They are different devices with different wiring and different resolution, and confusing them is a classic reason a motor will not start smoothly.
A debugging heuristic worth keeping
When an axis runs rough, a periodic torque variation that tracks the electrical angle is a typical clue that commutation, not the mechanics, is the source. Treat it as a clue, not a proof. Bearing friction, guideway drag, and load variation each produce their own periodic signatures, and you have to rule those out before you blame the feedback or the commutation offset.
Feedback becomes the ceiling on precision
With no gearbox to average out feedback error, the encoder is part of the axis, not a bolt-on. Your precision ceiling is whatever the feedback device can resolve and hold. Two decisions land straight in your firmware.
First, incremental or absolute. An incremental encoder sends counting pulses and needs a homing move after every power cycle. An absolute encoder assigns each position a unique code and reports true position the moment power returns, with no homing. On a direct drive axis that cannot afford a homing sequence at startup, absolute feedback is usually the default, and your boot code is simpler for it.
Second, resolution is not accuracy. A high step count per revolution tells you how finely the device subdivides, not how well it knows the true angle. Match the encoder to the axis and its error budget rather than to a headline number.
Time-grating encoders are one option worth knowing here. They are contactless and field-based, and they read angle from a phase shift in the time domain rather than by counting finely etched lines, so the precision comes from time-domain electronics. They are absolute, and on a direct drive axis they integrate as one feedback set with the motor. The honest boundary matters too. This is not the pick at the top metrology tier, where wafer-grade stages still run optical. It earns its place on large, hollow-shaft, or contaminated axes, where coolant and swarf would force an optical scale into a sealed, clean gap.
When direct drive is the wrong call
Direct drive is not free. The motor is larger, costs more at a given torque, and usually needs cooling. If your drawing tolerates backlash and you do not need the stiffness or the settling time, a geared axis is cheaper and simpler to build. The deciding question is usually blunt. Does a specification on your drawing fail if there is backlash? If nothing fails, you may not need a direct drive axis at all.
We build these. HansMotor has designed and built direct drive motors, linear motors, torque motors, and time-grating encoders in-house in Shenzhen since 2005, holds more than 230 patents, and is part of the Han's Laser group (SZSE: 002008). We also contributed to China's GB/T 34115-2017 standard for permanent-magnet linear motors. For the longer breakdown of when direct drive earns its cost, HansMotor has a full write-up.
Written by the HansMotor Engineering Team.
Top comments (0)