Function and importance of encoder in CNC machine tools

• 1
  What an encoder does inside a CNC machine tool
• 2
  Incremental, absolute and how the count is built
• 3
  Rotary and linear encoders in CNC machine tools
• 4
  Where encoder error fits in the total error budget
• 5
  When the encoder is not the weak link
• 6
  Encoder type compared

An encoder is a sensor that converts mechanical position or movement into an electrical signal. On a CNC machine tool it sits on the motor shaft or on the axis itself and reports where the axis is right now. The control compares that number with the commanded position and adjusts the drive output until the error falls inside the allowed window. That comparison runs thousands of times per second.

Two things come out of the signal: position and direction. Both matter. Direction tells the system whether the axis is moving toward or away from the target, which sets the sign of the correction. Without direction, a control cannot tell a 0.01 mm overshoot from a 0.01 mm undershoot.

The loop closes in the drive, not in the operator's head. A servo drive reads the encoder, subtracts the feedback value from the command, and pushes current into the motor in proportion to that error. Gains decide how hard it pushes. Too much gain and the axis hums or rings. Too little and it lags behind the toolpath on every corner.

Glass scales and motor-mounted encoders answer different questions. A motor encoder sees rotor angle, so it includes backlash, screw pitch error and thermal growth of the ball screw in the number it reports. A linear scale measures the table itself and removes most of that stack. That is why a machine with linear scales holds size better during a long warm-up run.

An incremental encoder emits pulses as the shaft turns. Electronics count the pulses and multiply by the distance per pulse to get position. On power-up the count starts at zero, so the machine must move to a reference switch or index mark before it knows where it is. That is the homing routine you see at the start of every shift.

An absolute encoder reports a unique code for every shaft position, so the control knows the axis location the moment power comes on. No homing move is needed. Multi-turn versions also track how many full revolutions have passed, which matters on a rotary table or a tool changer that spins many times.

Quadrature output is the common industrial format for incremental units. Two channels, A and B, sit 90 electrical degrees apart, and the order in which they switch tells the direction. A third channel, Z or index, gives one pulse per revolution for a reference point.

Resolution is not the same as accuracy. A 20-bit encoder divides one revolution into about 1,048,576 counts, yet the axis may still be off by 5 μm because of screw error. Counting finer only helps if the mechanical stack can hold that fineness under load.

Rotary encoders mount on the motor or the screw end. They are compact, sealed, and cheap to replace. Most 3-axis mills and lathes run this way. Their weak point is that everything between the encoder and the workpiece sits inside the measurement: coupling wind-up, thrust bearing play, screw stretch under load.

Linear encoders mount on the slide and read a scale fixed to the machine base. The table position is measured directly, so thermal drift of the screw drops out of the loop. Machines that hold ±0.005 mm over a long run usually need them. The trade-off is cost, mounting space, and a scale that must be protected from chips and coolant.

Ring encoders on a rotary table do the same job in rotation. On a Ø400 mm rotary table, an angle error of 5 arc-seconds is roughly 0.005 mm at the edge of the table. If the part features sit near the rim, that error is real and shows up in the inspection report.

Magnetic encoders are another option. They resist dust and moisture better than optical units but usually offer lower resolution. For a 3-axis machine cutting to ±0.05 mm, a magnetic unit is often enough. For grinding or jig boring, it is not.

Axis error comes from several sources stacked together: encoder resolution, screw pitch error, thermal growth, servo following error, and structural deflection under cutting force. The encoder is only one line in that list, and often not the largest.

A ball screw with a 10 mm pitch and a 2,500-line encoder in 4× quadrature gives 10,000 counts per revolution, or about 0.001 mm per count. That sounds better than the machine can hold. Real accuracy is set by pitch compensation tables, bearing preload and how warm the machine is.

Thermal growth is the quiet one. A 1,000 mm steel screw grows about 0.012 mm per °C. Ten degrees of warm-up moves the axis 0.12 mm. A motor encoder never sees this. A linear scale does, and the control corrects for it automatically.

Following error is the gap between commanded and actual position while the axis is moving. It grows with feed rate and drops with higher gain and stiffer mechanics. On a finishing pass at 200 mm/min the gap is small. On a roughing pass at 5,000 mm/min it can reach tenths of a millimeter, and the control has to slow into corners to keep it in check.

Fitting a higher-count encoder rarely fixes a size problem on its own. If the machine cannot hold ±0.02 mm and the fault is ball screw backlash or a worn thrust bearing, a finer scale just reports the same error more precisely. Measure first, then buy.

Backlash shows up as a step in the part when the axis reverses. Cut a square pocket and measure the wall positions on the two sides. A difference of 0.03 mm on the Y walls points to backlash, not to the encoder.

Vibration and surface finish problems usually come from servo tuning, tool runout or workholding. Chattered walls at 1,200 Hz are structural. An encoder fault produces different symptoms: position alarms, following-error trips, or a single axis that drifts when the machine is idle.

Temperature matters more than most shops expect. A machine that holds size in January can drift in July. If the parts are small and the tolerance is tight, cut the first article, let the machine run 30 minutes, then cut a second one and compare. The delta tells you how much of the error is thermal.