Introduction to Sensors on the CNC System
A machine tool only moves as well as it can measure. This page explains what each sensor on the CNC system actually reads, how the readings reach the control loop, and which ones you should pay for on a given part. Written for engineers and buyers who need to judge a machine or a quote, not a brochure.

In this article
- 1
- 2
- 3
- 4
- 5
- 6
How sensors on the CNC system close the loop
Every CNC machine runs the same basic loop. The control generates a commanded position, the drive moves the axis, a sensor reports the actual position, and the difference between the two becomes the next correction. Sensors on the CNC system are what make that loop real. Without feedback, the control is just sending pulses into the dark.
Two feedback families cover most machines. Semi-closed feedback reads the motor shaft through an encoder and assumes the ball screw and nut transmit that motion faithfully. Closed feedback reads a linear scale mounted on the slide itself, so screw pitch error, thermal growth of the screw, and backlash drop out of the reading. The trade is cost and mounting space.
The distinction matters when you read a tolerance on a quote. A machine with semi-closed feedback can hold ±0.005 mm on a good day, but it needs warm-up and regular pitch compensation. A machine with linear scales holds the same number across a long shift with less drift. Ask which one you are buying.
Feedback alone does not make a good part. The sensor tells the control where the axis is, not where the cutter edge is. Tool length, tool wear, spindle thermal growth and workpiece clamping all sit outside that loop until you add sensors that measure them directly.
- 1Semi-closedEncoder on the motor; screw errors stay in the loop.
- 2ClosedLinear scale on the slide; screw errors leave the loop.
- 3Resolution ≠ accuracyA 0.1 μm count does not mean a 0.1 μm part.
Encoders and linear scales: what the numbers mean
A rotary encoder turns shaft rotation into a pulse train. Incremental types give counts from a zero mark, so the machine must home after every power cycle. Absolute types know their position immediately and are the reason modern machines start without a homing routine. For a lathe C-axis or a rotary table, absolute feedback saves setup minutes every day.
Linear scales work on the same optical or magnetic principle but mount along the axis. Glass scales are the common choice on machining centers and reach resolutions around 0.1 μm. Magnetic tape scales tolerate coolant and chips better, cost less, and typically land near 1 μm. Neither number describes the machine's real accuracy, because the scale only reports where the slide is.
Resolution and accuracy are different quantities. Resolution is the smallest step the sensor can report. Accuracy is how close that report is to the truth. A scale with 0.1 μm resolution and ±3 μm accuracy over 1 m will hold a straight line to ±3 μm, not ±0.1 μm. Repeatability is a third number, and for most production work it matters more than both.
When you compare machines, ask for the accuracy grade of the scale, not the count. On our 5-axis centers we work to ±0.005 mm (±0.0002 in) and verify with 100% inspection before shipment, because the scale grade and the inspection routine have to agree.
- 1IncrementalNeeds homing; cheaper per axis.
- 2AbsoluteKnows position at power-up; faster setup.
- 3Glass vs magneticGlass for resolution, magnetic for dirty environments.
Temperature sensors and why thermal drift is the real limit
Steel grows about 11 μm per meter per °C. Aluminium grows roughly twice that. A 1 m aluminum part that warms 5 °C during roughing has moved more than 0.1 mm before any cutting error is counted. No encoder can correct this, because the scale is mounted on the machine, not on the part.
That is why temperature sensors matter. Thermocouples and PT100 elements sit on the spindle housing, the ball screw, the bed and sometimes the workpiece. The control uses those readings to compensate screw growth and spindle extension in real time. Machines without compensation rely on warm-up cycles instead.
Warm-up is not superstition. A spindle that has run for 40 minutes is a different size than a cold one. If your first part is inspected cold and your last part is inspected hot, the two reports will disagree even when the machine is behaving. For tight work, run a warm-up cycle and keep the shop temperature stable.
Practical rule: if your tolerance is tighter than ±0.02 mm on a part longer than 300 mm, ask how the machine handles thermal growth. Compensation sensors, a temperature-controlled room, or both. On a 4,000 mm part, thermal control is not optional.
- 1Spindle growthMain driver of Z-axis drift during a shift.
- 2Screw growthChanges pitch over the axis length.
- 3Part growthSensor compensation cannot see it.
Force, vibration and acoustic sensors on the cutting path
Position sensors describe the machine. Process sensors describe the cut. Spindle load monitoring is the simplest form: the drive current rises with cutting force, so the control can watch for a dull tool, a deep pass, or a missing tool. Many controls already log spindle load, so this costs nothing extra to use.
Accelerometers and vibration sensors catch chatter before it marks the surface. A microphone or acoustic emission sensor listens to the cutting zone and flags the frequency band that appears when a tool starts to sing. On deep pockets in 17-4PH or titanium, that warning arrives before the finish does.
Force sensors in the tool holder or the table give the most direct reading and the highest cost. They are worth it when a broken tool is expensive. Drilling a 20 mm hole in Inconel with an unattended cycle is a good candidate. Cutting 6061 on a short cycle with an operator standing there is not.
These sensors do not improve the part by themselves. Someone has to set the threshold, and the threshold has to be loose enough to avoid false stops. A monitoring system that trips every hour gets switched off within a week.
- 1Spindle loadAlready in most controls; free to use.
- 2VibrationDetects chatter and tool imbalance.
- 3ForceBest for unattended, high-value cuts.
Touch probes and tool setters: measuring without unclamping
A touch probe is a kinematic switch on a stylus. When the stylus contacts a surface, the switch closes and the control records the axis position. That single event gives you workpiece location, feature size, or a datum to shift the work offset. The accuracy is roughly ±1 to ±2 μm on a good probe, plus whatever the machine contributes.
Tool setters do the same job for the tool. A probe or laser sits off the table, and the control touches each tool to record length and diameter. This removes manual offsets and catches a chipped edge before the first cut. On a 20-tool job, that is minutes saved and one less scrap part.
In-process probing on the machine is not a substitute for a CMM. Coolant film, chip contact and stylus bending all add error, and the machine is not at 20 °C. What probing does well is establish datums and catch gross errors early, so the part that reaches final inspection has a chance.
We use probing to set work offsets on five-axis parts where re-clamping would introduce its own error. Final dimensional reports still come from inspection, and we share them on request.
- 1Workpiece probeSets datum and finds stock position.
- 2Tool setterRecords length and diameter automatically.
- 3Not a CMMGood for datums, not for final sign-off.
Which sensor does what
Match the sensor to the error you are trying to remove.
| Sensor | What it measures | Typical resolution | Use it when |
|---|---|---|---|
| Rotary encoder | Motor shaft angle | 0.1–1 μm equivalent | General positioning, most 3-axis work |
| Linear scale | Slide position directly | 0.1–1 μm | Tolerances at or below ±0.01 mm |
| Temperature probe | Spindle, screw, bed | 0.1 °C | Long parts, tight tolerance, long shifts |
| Spindle load | Cutting force proxy | Percent of rated load | Tool wear and breakage detection |
| Vibration sensor | Chatter and imbalance | Frequency domain | Deep pockets, thin walls, hard alloys |
| Touch probe | Workpiece datum and size | ±1–2 μm | Five-axis setup, first-part checking |
| Tool setter | Tool length and diameter | ±1–2 μm | Multi-tool jobs, unattended runs |
Where to spend first
If your tolerance is looser than ±0.05 mm, spend on a tool setter and spindle load monitoring; they cut scrap and setup time. If you need ±0.005 mm or better on long parts, spend on linear scales and thermal compensation first, because nothing else will hold that number through a shift.
Questions engineers ask
Does a higher-resolution encoder make my part more accurate?
No. Resolution is the smallest step the sensor reports; accuracy is how close that report is to the true position. A 0.1 μm encoder on a screw with 5 μm of pitch error still produces a 5 μm error.
Look at the whole chain: screw grade, scale accuracy, thermal behavior, and the inspection routine that confirms the result.
When is a linear scale worth the extra cost?
When the tolerance is at or below ±0.01 mm, when the part is long enough for screw growth to matter, or when the machine runs unattended for hours. Below those conditions a good encoder and regular pitch compensation usually hold the number.
On a 4,000 mm part, the answer is almost always yes.
Can in-process probing replace final inspection?
Not for tight work. The probe reads through coolant film and chips, the stylus bends slightly, and the machine is not at metrology temperature. Its real value is setting datums and catching a wrong offset before the part is finished.
Final dimensional reports should still come from a controlled inspection step.
Why does the first part of the shift measure differently from the last?
Thermal growth. The spindle and ball screws warm up as the machine runs, so the geometry changes by tens of microns over a few hours. Machines with temperature compensation adjust for it; machines without it need a warm-up cycle.
Keeping the shop temperature stable does as much as any sensor.
Do force sensors make sense on every job?
No. They cost more and need tuning. They pay off when a broken tool is expensive, when the cut is unattended, or when the material is hard enough to snap an edge without warning.
For short cycles in aluminum with an operator nearby, spindle load monitoring is usually enough.
What should I ask a supplier about their machine sensors?
Ask three things: feedback type (encoder or linear scale), whether thermal compensation is active, and how the final dimensions are verified. The answers tell you more than a tolerance line on a quote.
We run 100% inspection before shipment and share reports on request.
Send us the drawing and the tolerance
Upload your files and we will return a quote with free DFM analysis within 12 hours. Production can start within 24 hours, and parts ship in 3–5 days.
12-hour quote100% inspectionNo minimum order quantity