What Is an Encoder in CNC Machine Feedback?

• 1
  What an encoder in CNC machine control actually does
• 2
  How the pulse train becomes a position number
• 3
  Incremental, absolute, and linear encoders
• 4
  What encoders cannot fix
• 5
  Matching feedback to the parts you quote
• 6
  Symptoms that point at the feedback loop
• 7
  Encoder type compared by job requirement

An encoder is a sensor that reports motion. On a CNC machine it is bolted to a motor shaft, a ball screw end, or a linear scale rail. As that shaft turns or that rail slides, the encoder sends a stream of pulses back to the drive or control. The control compares the commanded position with the reported position and corrects the difference. That comparison is the whole point.

The controller issues a move, for example 100.000 mm on X. Without feedback, the machine assumes the motor did what it was told. With feedback, the control knows the axis reached 99.987 mm and can command the remaining 0.013 mm. No feedback, no correction.

This is why the term closed loop matters. An open-loop machine counts motor steps and hopes for the best. A closed-loop machine measures and adjusts. The encoder in a CNC machine is the measuring half of that loop, and it runs thousands of times per second during a cut.

Where does the signal go? Usually to the servo drive first, then to the CNC. The drive closes a tight velocity loop, the CNC closes the wider position loop. Two loops, one sensor. If the sensor lies, both loops chase a false number and the part comes out wrong.

Most rotary encoders use a disc with fine radial lines and an optical emitter-detector pair. Light passes through the lines and is chopped into pulses. A photodetector converts that flicker into a square wave. Count the pulses and you have counted the rotation.

Two channels, A and B, sit 90 degrees out of phase. That phase offset is not decoration. When A leads B, the shaft turns one way. When B leads A, it turns the other way. Direction comes free with the same two wires.

A third channel, Z or index, fires once per revolution. It gives an absolute reference point so the control can find home after power-up. Without Z, the control would only know how far it moved, never where it started.

Resolution is the angle or distance represented by one pulse. A 10,000 line encoder on a 10 mm pitch ball screw gives roughly 0.001 mm per count before quadrature. Quadrature reads all four edges of the A/B pair, so the effective count quadruples. Finer counts help, but they do not fix a loose coupling or a bent screw.

Incremental encoders report change, not location. Power down and the count is gone. On restart the machine must home to a reference, which costs time and depends on a switch or a Z pulse being reliable. They are cheap, fast, and everywhere on older mills.

Absolute encoders report a unique code for every shaft position. The control knows the axis location the moment power returns. No homing, no lost position after an e-stop, no scrapped part because someone jogged the axis while the control was off. Absolute feedback costs more and needs a serial protocol, not just two wires.

Linear encoders mount on the machine structure and read a scale instead of a shaft. They measure the table itself, so they see screw pitch error, thermal growth, and backlash. Rotary feedback cannot see those errors because it watches the motor, not the part.

The practical split: rotary for general milling and turning, linear on jobs where the tolerance is tight and the part is long. A 4,000 mm part with a ±0.005 mm callout is a linear scale job. A bracket with a ±0.05 mm callout is not.

An encoder measures position. It does not measure cutting force, tool wear, or thermal drift in the spindle. If the tool pushes off by 0.02 mm under load, the encoder still reports that the axis is exactly where it was told to be. The error lives in the structure, not the loop.

Resolution is not accuracy. A scale that counts to 0.0001 mm can still sit 0.01 mm off if it is mounted on a machine that flexes. Buyers often read the resolution number and stop there. That number describes the smallest step the control can see, not the largest error the machine can produce.

Encoder failure modes are usually mechanical or environmental. Contaminated optics from coolant mist, a loose coupling, a cracked glass disc after a crash, a cable shield that picks up drive noise. Symptoms show up as drift, following error alarms, or a surface finish that changes mid-cut.

When a machine holds ±0.005 mm all day, the encoder is one link in a chain that also includes the screw, the bearings, the spindle, and the room temperature. Fixing the feedback alone rarely fixes the part.

Start from the tolerance, not the encoder spec sheet. A part held to ±0.05 mm on a 200 mm aluminum bracket does not need linear scales. Rotary feedback on a good ball screw is enough, and it is faster to set up.

Move to linear scales when the part is long, when the tolerance is under about ±0.01 mm, or when the material cuts hot. Aluminum engine blocks and long titanium ribs both fall into that zone. So do parts that get measured in a temperature-controlled room after machining.

Absolute feedback pays off on unattended or lights-out runs. If a machine sits idle overnight and restarts without homing, a crash from a lost position disappears. For high-mix shops running 24 hours, that alone can justify the cost.

Ask what the machine will be asked to do in three years. Feedback upgrades are possible but rarely cheap. Buying the right loop at the start costs less than adding scales to a machine that was never prepared for them.

Following error alarms during acceleration usually mean the loop cannot keep up. Check the coupling first, then the cable shield, then the drive tuning. A loose coupling shows up as a small oscillation that grows with speed.

Drift that appears after two hours of cutting is often thermal, not electrical. The encoder is reporting correctly; the screw has grown. Linear scales on the table would catch that, rotary feedback on the motor end cannot.

A single axis that finishes rough while its neighbors cut clean is a strong clue. Swap the encoder cable if the machine allows it. If the problem follows the cable, you found it. If it follows the axis, look at the mechanics.

Repeatability checks with a dial indicator still matter. Command a move, measure the result, repeat ten times. If the spread is wider than the spec, the loop, the screw, or the thrust bearing is the suspect, in that order.