CNC Encoder: Precision Feedback Tool
A CNC encoder is the feedback loop that tells the controller where an axis actually is, not where it was told to go. This page covers how incremental and absolute feedback work, which specifications decide machined accuracy, and the cases where encoder choice stops being the limiting factor.
In this article
- 1
- 2
- 3
- 4
- 5
- 6
What a CNC encoder precision feedback tool actually does
A closed-loop axis has two signals: the command the controller sends to the servo drive and the position the encoder reports back. The drive compares them thousands of times per second and adjusts current to the motor until the error falls inside the loop bandwidth. Without that comparison, the machine only knows how long it commanded the motor to run, not where the slide ended up.
Cutting forces push back. A Ø16 mm end mill in 4140 steel at 0.08 mm per tooth draws enough load to deflect the tool, the holder and the column. Thermal growth moves a 4,000 mm machine bed by more than the drawing tolerance over a long run. An encoder sees the resulting position error and the controller compensates before the next tooth engages.
Feedback resolution is not the same as feedback accuracy. Resolution is the smallest position step the encoder can report. Accuracy is how close that reported step is to true position. A 1 μm resolution scale mounted on a bracket that flexes 10 μm gives you a very fine reading of the wrong number.
The encoder also closes the loop on the spindle in many machines, which is how rigid tapping and oriented tool changes stay repeatable. On a mill-turn center, a separate encoder on the C-axis lets the controller interpolate turning and milling in one setup.
Incremental vs absolute: which feedback suits which machine
An incremental encoder outputs pulses as the axis moves. The controller counts edges from a home reference to know position. Resolution follows the line count on the disc or scale, and quadrature decoding multiplies the count by four. Incremental feedback is simple, fast and inexpensive. Its weak point is the homing move after every power cycle, and position is lost if a pulse train is disturbed or the drive faults mid-cut.
An absolute encoder reports a unique position value for every point on its travel. A single-turn absolute reads position within one revolution; a multi-turn version also counts shaft rotations, so the controller knows the slide position at power-up with no homing. That matters on a five-axis machine where a tilted head must be re-datumed safely, and on any cell where a robot or pallet changer loads parts without an operator present.
Linear scales are the third option. They mount on the machine structure and read the slide directly, so they include ballscrew pitch error, thermal growth of the screw and backlash. Rotary encoders on the motor shaft cannot see any of those. For a part held to ±0.005 mm over 750 mm of travel, that difference shows up in the Cpk.
Absolute linear feedback costs more per axis and needs a clean mounting surface. Incremental rotary feedback on a ballscrew is the usual choice for general milling and turning where the machine is homed at the start of a shift.
Reading encoder specifications against a real tolerance
Resolution is quoted in counts per revolution or in μm for a scale. Take a 20 mm pitch ballscrew with a 10,000 count per revolution encoder: one count equals 2 μm of slide travel before any electronic interpolation. Interpolation can push the reported number finer, but it does not create new information about true position.
System accuracy is the number that matters. It combines encoder accuracy, scale mounting, screw pitch error, thermal effects and servo stiffness. Published encoder accuracy for a good linear scale sits in the ±2 to ±5 μm per meter class. Machining tolerance for tight work here is ±0.005 mm, so the feedback chain has to be a small share of the budget, not most of it.
Repeatability and accuracy are separate budgets too. A machine that returns to the same point within 2 μm but sits 15 μm off nominal is repeatable and inaccurate. Encoder feedback helps with repeatability directly and with accuracy only when the scale is referenced to the part datum, not to a bracket that moves with temperature.
Signal quality sets a ceiling as well. Differential line drivers, shielded cable and separation from spindle and drive power wiring keep the pulse train clean. Noise on an incremental channel looks exactly like motion to the controller, and no amount of tuning fixes a corrupted count.
When the encoder is not the limiting factor
Feedback cannot correct a machine that is not stiff enough. If the column deflects 30 μm under a heavy radial cut, the encoder reads the deflection and the controller tries to push back, which usually shows up as chatter rather than better size. The fix is a lighter radial engagement, a shorter tool, or a different setup.
Feedback cannot fix thermal drift that the controller does not model. A machine running unattended for six hours will grow. Scales that read the slide directly remove screw growth from the equation, but the part and fixture grow as well. On long runs, in-process gauging or a mid-run re-datum does more than a finer encoder.
Very high resolution also raises tuning demands. Servo gains, filters and feed-forward must match the feedback, or the axis will buzz at standstill and leave faceting on contoured surfaces. Fine feedback on a lightly damped axis is a recipe for audible oscillation at 100 to 300 Hz.
Tool wear is outside the loop entirely. The encoder knows where the slide is, not how much the insert has worn. A 0.02 mm flank wear on a turning insert shows up in the diameter no matter how good the feedback is.
How we hold tolerance on encoder-driven axes
We run 127 high-precision CNC machines across three plants, including 16 simultaneous 5-axis machining centers with Ø400 mm rotary tables and a maximum processing size of 4,000 mm. The five-axis work is where feedback choice shows up most, because a tilted B-axis error is amplified by the distance from the rotary center to the cutting edge.
For parts held to ±0.005 mm and finishes between Ra 0.2 μm and Ra 1.6 μm, we match the machine to the feature. A deep rib in a 7075 pocket on a 5-axis center with rotary feedback behaves differently from the same rib cut on a three-axis machine with a long reach tool. The tolerance is the same; the error sources are not.
Setup on a used machine is where most feedback problems appear. A slightly loose coupling between motor and screw, or a scale head air gap out of specification, will produce a periodic error that repeats once per revolution. On a turned diameter that reads as lobing; on a milled face it reads as a repeating step.
We inspect 100% before shipment, with raw material checks, in-process monitoring and final reports on request. That catches the part-level result, but the feedback chain is what keeps the process repeatable between inspections. Materials we machine daily include 6061-T6, 304 stainless, 17-4PH, TC4 titanium and POM.
Encoder feedback types compared
Match the feedback type to machine class, tolerance and environment.
| Feedback type | Typical resolution | Best for | Watch out for |
|---|---|---|---|
| Incremental rotary on screw | 1–5 μm at the slide | General 3-axis milling and turning | Homing after every power cycle |
| Absolute rotary multi-turn | 1–5 μm at the slide | 5-axis and automated cells | Higher cost per axis |
| Incremental linear scale | 0.1–1 μm on the scale | Tight tolerance over long travel | Clean mounting surface required |
| Absolute linear scale | 0.1–1 μm on the scale | Unattended, high-value parts | Cable routing and cost |
| Magnetic encoder | 5–20 μm at the slide | Dirty, wet or vibrating axes | Lower accuracy than optical |
| Spindle encoder | Depends on drive, often 1:1 | Rigid tapping, oriented stops | Belt or coupling slip |
Which feedback to specify
If the part needs ±0.005 mm over long travel or runs unattended, specify absolute linear scales on the critical axes. If the work is general milling and turning on a homed machine, incremental rotary feedback on the ballscrew is enough and costs far less.
Encoder feedback questions engineers ask
Does a higher count encoder automatically improve part accuracy?
No. Resolution sets how finely position is reported, not how close the reading is to true position. Once resolution is roughly one tenth of the tolerance you need, the remaining error usually comes from screw pitch, thermal growth, machine stiffness and tool wear.
A 1 μm scale on a flexing bracket reports the flex very precisely. Fix the structure before buying more counts.
Why does an axis lose position after a power cut?
Incremental encoders count pulses from a reference, so position is gone when power drops. The controller must re-home before cutting. Absolute encoders keep a unique value per position and are ready at power-up, sometimes with a battery or an energy-harvesting multi-turn counter.
If your cell restarts unattended, absolute feedback removes the homing move and the risk of a crash during re-datum.
Can encoder feedback compensate for thermal growth?
Partly. Linear scales read the slide directly and remove ballscrew growth from the loop. They do not remove growth of the part, fixture or column. Controllers with thermal models use spindle and ambient sensors to offset the rest.
On a six-hour unattended run, a mid-run re-datum or in-process gauging is often more effective than a finer encoder.
What causes a periodic error that repeats once per revolution?
Usually a mechanical coupling issue: a loose or misaligned motor-to-screw coupling, a bent screw, or a scale head air gap outside its specification. The error repeats at the rotation frequency because the cause rotates with the shaft.
Check coupling runout and scale gap first. Re-tuning the servo will not remove a mechanical once-per-rev error.
Is an encoder the same as a linear scale?
No. A rotary encoder reads motor or screw rotation; a linear scale reads slide position directly. A scale therefore includes screw pitch error, backlash and screw thermal growth in its reading, which a rotary encoder cannot see.
For tight tolerances over long travel, that difference is often larger than the encoder resolution itself.
How much of a machining tolerance should feedback use?
A practical rule is to keep the whole feedback and motion chain within one third of the tolerance, then leave the rest for thermal, tool and fixturing effects. For ±0.005 mm work that is a tight budget.
When the budget does not close, change the process: fewer setups, shorter tools, or in-process measurement.
Send the drawing, get a real process answer
Upload your files and we return a quotation with free DFM analysis within 12 hours. Production can start within 24 hours, and uploads stay secure under NDA on request.
12-hour quoteNo minimum order quantity100% inspection before shipment
More from the shop floor
We publish setup notes, tooling trials and inspection data from the factory floor.
