The three numbers
Servo sizing has a reputation for spreadsheets, but the first pass needs three numbers and a catalogue:
- Continuous torque the load demands, including friction and gravity, with a margin for voltage variation.
- Peak torque during acceleration, which is what usually selects the frame size rather than the continuous figure.
- Speed at the load, after the gearbox or the pulley ratio — not the motor's rated speed.
Everything else — encoder resolution, bandwidth, tuning — refines an axis that already works. Get these three wrong and no amount of tuning saves it.
Inertia ratio is a tuning constraint, not a rule
The inertia ratio (load inertia divided by rotor inertia) predicts how hard the axis will be to tune. A ratio under 5:1 is comfortable, 10:1 is common on a belt axis, and beyond 20:1 the axis will move but the settling time and the overshoot become the application's problem rather than the drive's.
The practical lever is the gearbox or the pulley ratio: load inertia reflected through a reduction falls with the square of the ratio, so 3:1 gearing cuts reflected inertia by nine. That is usually cheaper than the next frame size up.
The R8 and S8 drives both identify inertia online and offline, which shortens commissioning, but identification measures a ratio that is already there — it does not fix a bad mechanical choice.
Encoder choice: bits, and turns
Two independent decisions. Bits: 17-bit gives 131,072 counts per revolution, 23-bit gives 8,388,608. For a positioning axis with a fine pitch screw, 23-bit buys resolution that no amount of electronic gearing can recreate from 17-bit feedback. For a conveyor or a fan, 17-bit is ample.
Turns: single-turn encoders report position within one revolution and need no battery. Multi-turn encoders retain absolute position over 65,535 revolutions, which removes the homing cycle — and requires a battery in the six-core encoder cable. That battery is a maintenance item, and it belongs in the machine's service schedule.
Pulse, analogue or bus
Pulse command is the simplest and the most portable: 500 kpps differential on the R8, with electronic gearing up to 1:1,073,741,824. It suits retrofits where the existing controller already outputs pulses, and axes that do not need tight coordination.
EtherCAT is the answer for coordinated motion: 2×100 Mbps, 100 axes at 0.1 ms of bus delay, a 1 ms cycle as standard and integer multiples of 250 µs below that, with CSP, CSV and CST modes. Once a machine interpolates between axes, the bus stops being a luxury.
Analogue ±10 V still appears on legacy machines. It works, it drifts, and it needs shielding — specify it only where the controller cannot do anything else.
Cables are part of the axis
Standard lengths are cheaper and stock faster. Beyond 15 m the cores get thicker and the price rises noticeably, which surprises people who sized the motor carefully and then mounted the cabinet at the far end of the machine.
Drag-chain cable is rated for 5 million bending cycles by default; a static run does not need it, and a moving axis should not use anything else. The brake cable is separate from the power cable below 80 flange, and the encoder cable carries the battery on multi-turn motors — three cable part numbers per axis, not one.