Performance of the CNC machine servomotor system during use
A CNC machine servomotor system does not hold position on its own. It trades loop gain for stability, and resolution for speed. This page explains what you can measure on the shop floor, which numbers matter, and when a servo problem is really a mechanical one.

What a CNC machine servomotor system actually controls
A CNC machine servomotor system is a closed loop. The control sends a position command, the drive converts it into torque, the motor turns a ball screw or a pinion, and a feedback device reports where the axis really went. Every error you see on a part traces back to one link in that chain.
The loop does not care about the drawing. It only compares command and feedback. If the feedback scale sits on the motor shaft, screw pitch error, thermal growth, and backlash never enter the loop. The table can drift and the drive will still report near-zero error.
That is why machine builders put linear scales on large and medium machines. On a 4,000 mm travel, a 0.01 mm pitch error over the screw length is larger than the tolerance band itself. Direct measurement of the slide removes that term from the error budget.
Open-loop gain K sets how hard the loop pushes back. Typical CNC servo mechanisms run K between 20 and 30 1/s. Below 20, the axis feels soft on direction changes and contour error grows on arcs. Push much above 30 and the axis starts to ring.
- 1Feedback location decides what is controllableMotor-mounted encoder cannot see screw or thermal error.
- 2Axis inertia sets the ceilingHeavy tables need more gain, not more motor.
- 3Resolution is not accuracyA fine scale can still read a loose axis.
Loop gain, damping, and step response in use
In a second-order model, the damping coefficient falls as gain rises. A step command that overshoots or oscillates means the loop has more gain than the mechanics can absorb. The usual fix is not to lower K but to stiffen the drive train.
A servo system with K below 20 is generated by a soft coupling, a worn ball screw, or an oversized load on a small motor. The drive cannot correct a mechanical spring. Gain tuning only exposes it.
Watch the following error during a slow feed. On a healthy axis it stays constant and small. If it climbs as the axis heats up, the problem is thermal, not electrical. If it jumps at one spot on the travel, look for a tight spot or a damaged screw.
The step response test still tells you the most in the least time. Command a small move, capture the encoder trace, and read overshoot, rise time, and settling. No proprietary tool needed. A scope and a drive trace are enough.
- 1Constant following errorNormal. It scales with feed rate.
- 2Rising following errorThermal growth or a dragging slide.
- 3Spike at one positionMechanical defect on the screw or guide.
Resolution and feedback accuracy in the detection chain
Resolution is the smallest displacement the system can report. It is a property of the detection element plus the measuring line, not of the motor. A drive rated at 1 μm resolution cannot resolve better than the scale feeding it.
Linear induction synchronizers used on larger CNC machines reach ±0.0001 mm accuracy, 0.05 μm sensitivity, and 0.2 μm repeatability. Those numbers describe the sensor. The machine around it rarely holds that well once load and heat are applied.
In a closed loop, the error from the detection element and the true axis deviation are hard to separate. If the scale reads high by 2 μm, the loop will drive the table 2 μm off and call it correct. Calibration of the feedback device matters as much as its resolution.
For most parts machined to ±0.005 mm, a standard glass scale is enough. Chasing 0.05 μm sensors on a machine with 0.5 μm of thermal drift per hour buys nothing measurable on the part.
- 1Match sensor to machineA 0.2 μm scale on a 5 μm frame is wasted.
- 2Repeatability beats resolutionRepeatable error can be compensated.
Speed against stability: where servo tuning goes wrong
High gain gives fast response and tight contouring. It also amplifies every resonance in the structure. On a machine with a long, thin ball screw, the screw's own axial resonance sits low enough that aggressive tuning will excite it.
The practical approach is to set gain as high as the mechanics allow, then back off slightly. A small amount of following error at high feed is cheaper than a surface that rings on every corner.
Feed-forward terms change the picture. They reduce following error without raising the loop gain, so the axis stays stable while tracking improves. On contoured parts this shows up as better corner accuracy at the same cutting speed.
Mechanical fixes still beat electrical ones. Preloaded ball nuts, a stiffer coupling, and a properly aligned guide remove the error at the source. The servo only has to hold what the mechanics deliver.
- 1Raise gain until it ringsThen back off 20–30% for margin.
- 2Use feed-forward firstTracking improves without instability.
- 3Fix mechanics before tuningGain hides backlash, it does not remove it.
Servo behavior and what each symptom points to
Measure on the machine, not in the manual.
| Observation | Likely cause | Check first |
|---|---|---|
| Following error grows with feed | Gain too low for the load | Drive gain setting, axis inertia |
| Overshoot on every step | Gain above mechanical limit | Coupling stiffness, screw preload |
| Error rises after 1 hour | Thermal growth of screw | Scale compensation, cooling |
| Error spikes at one position | Local screw or guide defect | Runout check on the screw |
| Ringing on corners only | Resonance at one frequency | Notch filter, feed-forward |
| Position drifts at rest | Feedback or drive offset | Scale calibration, encoder signal |
What to fix, in what order
If the axis is soft at low speed, fix the mechanics first: coupling, preload, alignment. If it tracks well but rings, lower gain or add feed-forward. Adding a finer scale to a loose axis changes the number on the display, not the part.
Common questions
Does a higher-resolution encoder improve part accuracy?
Only if the mechanics can hold that resolution under load. A 0.1 μm encoder on an axis with 2 μm of backlash will report position the table never reaches. The loop will correct toward a number that does not match the workpiece.
Fix backlash and thermal drift first. Then improve resolution, and re-check the part, not the display.
Why does my machine hold size in the morning and drift by afternoon?
That pattern points to thermal growth, not the servo. A ball screw expands as it warms, and the axis position changes along the travel. Machines with scale feedback on the slide see less of this than machines reading the motor shaft.
Let the machine warm up through a spindle warm-up cycle, and check whether the drift repeats on the same part feature. If it does, it is thermal.
Is a servo fault always an electrical problem?
No. Overcurrent and following-error faults often come from a tight guide, a dry way, or chips packed under the slide. The drive pushes harder to overcome the friction, draws more current, and trips.
Clean and lubricate the axis, then check current draw again before replacing a drive.
How much following error is acceptable?
It depends on the tolerance you are holding and the feed rate at which error is measured. On a well-tuned axis, following error at a given feed is repeatable, and the control can compensate for it.
What you cannot compensate is error that changes with temperature, load, or position. Chase that kind first.
Do we need a linear scale on every axis?
Not always. On short-travel axes with a good screw and stable temperature, motor-mounted feedback holds well. On long travels, or where the part tolerance is tight relative to screw pitch error, a scale on the slide is worth it.
The deciding question is simple: can the error between motor and table exceed your tolerance? If yes, measure the table.
Send us the part that keeps drifting
Tell us the feature that moves out of tolerance and the machine it runs on. We quote and return a DFM analysis within 12 hours.
12-hour quote100% inspection±0.005 mmNo minimum order