CNC Probing & In-Process Inspection: How On-Machine Probes Work and When They Pay

The classic rhythm of machining is: cut the part, take it out, and find out at the CMM or the inspection bench whether it is right. Probing breaks that rhythm. An on-machine probe turns the CNC machine itself into a measuring device, so the part can be checked while it is still clamped where it was machined — and corrected before it ever leaves the fixture. Done well, probing is the difference between discovering an error on the third part and preventing it on the first. Done badly, it is a source of confident wrongness: a probe that is trusted but uncalibrated will tell you, precisely and repeatably, the wrong number.

This guide is the reference for that tool and that discipline. It explains the two probe families and what each measures, why a touch probe is only as good as its calibration, the honest limits of asking a machine to measure itself, where probing sits in the process, and how it divides the work with the CMM rather than replacing it. It is the measurement-automation anchor of this library, and it leans on three companions: the spec-sheet guide explains the machine accuracy that probing inherits, the thermal-deformation guide describes the drift probing exists to catch, and the surface-finish and tolerance guide covers the measurement honesty every inspection assumes. Terms like probing, CMM, datum and tolerance are in the glossary.

The two probe families: the workpiece probe and the tool setter

On-machine probing divides into two families, and the single fact that separates them is what is being measured — the part or the tool.

The workpiece probe (the spindle probe, part probe, or touch probe) is carried by the machine itself — mounted in the spindle on a machining centre or on the turret of a turning machine — and it measures the workpiece. It touches the part’s edges, faces, bores and bosses to find where the part actually is, and from those touches the control can set the work offset (the part’s origin), align a part or a rotary axis, check that the correct part or fixture is loaded, and measure features after machining. Communication with the control is wireless — by infrared or radio — or by cable, and the probe is stored in the tool magazine and loaded like any other tool when a measuring cycle calls for it.

The tool setter (the tool probe) is fixed to the machine — on the table of a machining centre or in the turret area of a turning machine — and it measures the cutting tool. The machine presents each tool to the setter, which measures the tool’s length and diameter and automatically writes those values into the tool offset table, so every tool is set to its true size before it cuts rather than by an operator touching it off by hand. Two kinds exist: contact setters, which the tool touches like a workpiece probe in reverse, and non-contact laser setters, which measure the tool with a light beam as it passes through — able to check a tool’s full profile, detect chipped edges, and monitor runout without ever touching the edge.

	Workpiece (spindle) probe	Tool setter (tool probe)
Where it lives	Carried by the machine — spindle or turret	Fixed to the machine — table or turret area
What it measures	The part: edges, faces, bores, bosses, position	The tool: length, diameter, breakage, wear
What it is for	Set the work offset, align parts, find the true position of stock, verify features	Set tool offsets automatically, detect broken or worn tools before they cut
When it runs	Before and between cutting operations	Between tools, and before a tool is trusted to cut

Within each family there is an extra level of sophistication worth naming once. The standard touch probe takes a single point at each contact. Scanning probes — the same idea with a more sensitive mechanism — trace continuously across a surface and record many points, which is how freeform and sculpted geometry is captured on the machine. And for measuring deep holes or features parallel to the spindle axis, probes are built with stylus configurations — star probes with several styli, or side-reaching configurations — that let the machine reach geometry a single straight stylus cannot. The families are the frame; the stylus and the probe type are chosen by the geometry being measured.

How a touch probe actually measures: a switch and a calibration

The single most important thing to understand about a touch probe is that it is not a measuring instrument in the way a CMM is. A touch probe is a switch. It registers contact not when the stylus ball touches the surface but when the stylus is deflected far enough that the probe’s internal mechanism triggers and sends a signal. That threshold of deflection is tiny — but it is not zero, and it is not perfectly consistent. The probe’s trigger force varies slightly with the direction the stylus is pushed, an effect called lobing: a probe triggered from one approach direction fires at a slightly different deflection than the same probe triggered from another. The mechanical repeatability of a good touch probe is excellent — that is what the makers quote in their specifications — but the point where it actually fires depends on direction, speed and the wear and condition of the probe.

What makes this a solvable problem rather than a fatal flaw is calibration, and here is the fact that separates shops that trust their probes from shops that are fooled by them: the correction lives in the calibration, not in the probe. Before a probe is used for real work, it is driven against an artifact of known size — a calibration sphere or a ring gauge whose diameter has been measured and trusted. Comparing what the probe reports with what the artifact truly is yields the stylus deflection and the effective radius of the stylus ball at the moment of trigger; the control stores those values and applies them to every future touch. The probe itself never changes; the calibration tells the control how to interpret what the probe reports. A probe that has never been calibrated, or whose calibration is stale, is measuring its own trigger behaviour and calling it the part — a source of confident wrongness.

Calibration carries three disciplines that shops ignore at their peril:

Calibrate under the conditions you measure in. The probe fires differently at different approach speeds and from different directions, so a calibration taken at one feed rate does not hold for a measurement taken at another. The rule is simple: calibrate at the same feed rate, in the same measuring plane, and from the same approach directions that production probing will use. On machines whose rotary axes move the tool — swivel-head and articulated machines — the probe’s relationship to the axis changes with every head angle, so the calibration must be done for each probe angle actually used. On trunnion and rotary-table machines, where the part tilts but the spindle-mounted probe does not, the probe calibration holds across table positions.

The artifact must be truer than the work. A calibration sphere or ring gauge that is itself out of round, or whose certified size is wrong, quietly poisons every measurement made against it. The artifact is the reference chain’s anchor — it should be a gauge-grade artifact with a known, periodically verified size, kept clean and handled like the standard it is. Garbage in the ring gauge means garbage in every part the probe ever checks against that calibration.

Recalibrate on the events that invalidate the old numbers. The list is short and unambiguous: when the probe is first set up; whenever the stylus is changed — even for an identical stylus, because the new stylus has its own effective length; after any crash or suspected damage to the probe or stylus; after significant machine service or repair; after long downtime; and periodically, because the machine the probe rides on drifts with wear and temperature. Shops doing critical work also run a quick repeatability check — a handful of touches on the same sphere in the same direction, watching the spread — when they doubt the data. Calibration is not a one-time setup; it is a habit tied to the machine’s life.

The honest limits: what on-machine measurement cannot verify

Every probing guide leads with the benefits, and they are real — but the tool earns its place only when its limits are understood, because those limits decide what to trust it for. The central fact is the one the makers state plainly and shops forget: the machine is measuring itself. The probe rides the machine’s spindle and reports positions in the machine’s own coordinate system, so everything the probe measures is expressed through the machine’s geometric accuracy, its repeatability and its current thermal state. The best probe in the world cannot see farther than the machine it rides on. This has two consequences that structure everything below:

Size and position yes; form no. A probe on a machine can measure where a feature is and how big it is — bore diameters, boss positions, edge locations, distances between features — because those compare the probe’s report against the machine’s own axes. What it cannot do is verify the machine’s own geometry: if the machine has backlash, a bowed axis or a tilted spindle, the probe is blind to it, because it measures through that very error. Roundness, flatness, squareness and true geometric form are exactly what an on-machine probe cannot certify — the machine would be certifying its own straightness with its own hand. The shop-floor saying captures it: the probe is excellent at finding a bore’s position and poor at proving the bore is round, because the roundness question is really a question about the machine and the probe is the machine.

A rough screen, not a fine screen. Because the machine-measures-itself error floor sits above the CMM’s, on-machine probing is a superb coarse filter and a weak fine one. It reliably catches the errors that matter most in production — a wrong setup, a missed operation, a crashed tool, a part drifting off with thermal growth, stock in the wrong place — the gross failures that would otherwise produce scrap until someone inspects. It is far less reliable at the marginal judgement: whether a feature that measures a few microns out is truly out, when the machine’s own uncertainty is the same size as the tolerance being judged. The division follows: probing screens the process; the authoritative measurement judges the part.

Probing costs spindle time. Every touch is time the spindle is measuring rather than cutting, and a probing-heavy cycle is a longer cycle. The honest accounting — covered below — weighs that time against the setup time, the scrap and the inspection queue it removes. Shops that skip the accounting and add probing everywhere pay for it; shops that skip probing to save ten seconds pay for it worse.

Where probing sits in the process

With the families and the limits in place, the workflow writes itself. Probing operates at three moments around the cut, and each has its own job:

Before machining (pre-process): find the truth of the setup. The probe locates the part and the fixture in the machine’s world: it finds the workpiece’s edges and faces, sets the work offset automatically, aligns parts and rotary axes, and confirms the correct blank is loaded before the program runs. This is where probing pays its most reliable dividend, because it kills the two great setup costs at once: the operator time spent aligning and touching off by hand, and the transcription errors that manual offset entry invites. It also absorbs variation the programmer could not foresee — a casting that shifts, a weldment with rough stock, a saw cut that is not square — by finding where the part actually is rather than assuming where the fixture put it. And it is the shop-floor anti-error device: probing a part before cutting confirms it is the right part, in the right place, before a wrong assumption machines expensive material.

During machining (in-process): catch the drift while it can still be corrected. Between operations, the probe measures the state of the work: how much stock remains before a finishing pass, whether the tool has worn or broken, whether the feature just cut landed where the program promised. This is the closed-loop idea at its purest — measure, compare, correct: the probe checks a feature, the control compares it to the target, and if the tool has drifted it adjusts the offset and re-cuts — a mill-check-adjust loop that keeps a worn tool producing good parts instead of bad ones until someone changes it. In-process probing is the practical answer to the thermal drift that walks a machine through the morning, because a probe that measures a critical feature before the finishing pass sees the accumulated error and lets the control correct it rather than cutting on faith. And it is the non-negotiable foundation of unattended and lights-out running: a machine that runs without an operator is only safe to trust if it checks its own work, detects its own broken tools, and stops or corrects itself when the process drifts — probing is the sense of hearing that lets a dark machine be left alone.

After machining (post-process, on-machine): screen before it leaves. With the part still fixtured, a final probe routine rough-screens the critical features — a fast pass that separates the clearly good from the suspect. The clearly good parts move on; the suspects are pulled for the authoritative check. Done consistently, this screening catches the process failures that would otherwise travel to the inspection bench as a queue of bad parts, and it does it while the part is still in the machine where — if something is wrong — it can sometimes still be saved.

Probing and the CMM: a division of labour, not a rivalry

The recurring mistake in the probing conversation is treating the probe as a cheap CMM — a replacement to be bought instead of one. It is not. The two instruments answer different questions at different moments, and the shops that use on-machine probing best are the ones that stopped arguing about which one is better and assigned each its proper job:

	On-machine probing	The CMM
When	During and around machining, part still fixtured	After machining, part removed
Where	In the machine, in the shop environment	In a controlled inspection area, usually a stable temperature
Accuracy	Limited by the machine it rides on	An order of magnitude finer, in a controlled room
What it is good at	Finding the part, screening the process, catching drift and gross errors early, every part	Judging the part — critical dimensions, form and geometric tolerances — authoritatively
Its cost profile	Spindle time, but no queue and no waiting	No spindle time, but travel, queueing and a bottleneck

The CMM’s strengths are exactly the probe’s weaknesses: it measures in a controlled environment, to finer resolution, with instruments that do not ride on the machine being judged — which is why it, not the probe, is the authority for final acceptance, first-article inspection and the form tolerances the probe cannot certify. The probe’s strengths are exactly the CMM’s costs: it checks every part at the moment the error happens, while the part can still be corrected, without the part ever leaving the machine or joining an inspection queue. A probing routine on a machine tool is far faster and cheaper per part than moving the part to the CMM — but it is a screening measurement, not the referee.

The practical synthesis is the one the best shops run: use probing as the rough screen and the CMM as the fine screen. Probing watches the process in real time — setup verified, tools checked, drift corrected, every part screened for gross error before it leaves the machine. The CMM takes the first article, the critical or geometrically complex features, and the samples that confirm the process — and because probing has already filtered the obvious failures, the CMM’s sample rate can be far lower than it would be for a shop inspecting by CMM alone. The two instruments also meet in the middle on the work the CMM genuinely cannot do: very large parts that will not travel to an inspection room, and in-fixture correction of expensive workpieces that cannot be re-set up after a CMM verdict. For that class of work, on-machine measurement is not a second-best substitute — it is the only measurement that can feed a correction back to the same setup.

When probing pays — and when it does not

The return on probing comes from the failures and setup hours it removes, so the honest test is whether your work produces those failures and hours. Probing pays most when:

• Setups change often. High-mix work, small batches, frequent changeovers — every setup is where probing’s alignment, offset-setting and error-prevention earn their keep. A shop that runs long, stable, identical jobs has fewer setups for probing to save.
• Tolerances are tight enough that the process must be watched. When the tolerance is smaller than the comfortable margin of a fresh tool and a warm stable machine, probing is the instrument that tells you the process is still inside the window before you machine ten more parts outside it.
• Parts need several setups or several faces. Each re-fixture re-introduces error; probing re-locates the part by its actual position rather than trusting fixture repeatability, and checks the relationships between features machined in different setups.
• Drift is a live risk. Long runs, warm machines, worn tools — anything that walks the process over time. In-process probing is the closed-loop correction for the thermal and wear drift that no static program can anticipate.
• The work runs unattended. Lights-out, automation, robot-tended cells — an unmonitored machine needs probing to find its parts, check its tools, and catch its own failures. The lights-out guide treats probing as a precondition, not an option.
• Parts are large, expensive or variable. A big mold or aerospace part that cannot easily travel to a CMM, or a casting whose stock position varies — probing inspects where the part is and protects expensive material from being cut wrong.
• First-part and marginal-part risk matters. Rough-screening every part catches the process failure before it becomes a queue of scrap.

And the equally honest counter-signals: simple, stable, loose-tolerance three-axis work in long runs, where setup is rare and the tolerance window is wide, may never repay a probe package; a machine whose own accuracy and stability cannot be trusted gives its probe nothing to stand on; and a shop that installs probes but does not maintain the calibration discipline has bought a device for generating confident wrongness. The decision is a process decision, not a purchase decision — the probe is a tool the process either uses or does not.

Frequently asked questions

What is the difference between a workpiece probe and a tool setter? What they measure. The workpiece probe is carried by the machine — in the spindle or turret — and measures the part: it finds edges and faces, sets the work offset, aligns parts, and checks machined features. The tool setter is fixed to the machine and measures the cutting tool: it sets each tool’s length and diameter into the offsets automatically and detects broken or worn tools before they cut. One measures what you are making; the other measures what you are cutting with.

Why does a touch probe need calibration? Because a touch probe is a switch, not a ruler. It registers contact when the stylus is deflected far enough to trigger, and that trigger deflection varies slightly with direction and speed. Calibration drives the probe against an artifact of known size — a sphere or ring gauge — to measure the stylus deflection and effective ball radius, and stores them so the control can interpret every future touch correctly. The correction lives in the calibration, so a probe that is never recalibrated is measuring its own trigger behaviour and calling it the part.

How often should I calibrate my probe? Whenever an event invalidates the old calibration: first setup, any stylus change (even an identical stylus), a crash or suspected damage, significant machine service, or long downtime — and periodically, because the machine the probe rides on changes with wear and temperature. Calibrate under the same conditions you measure in — same feed rate, plane and approach directions — and recalibrate each probe angle actually used on machines whose head tilts. Critical-work shops also spot-check repeatability with a few touches on a sphere when they doubt the data.

Can on-machine probing replace a CMM? No — and the mistake is treating it as one. The probe rides the machine, so it measures through the machine’s own geometric and thermal errors: it is excellent at position and size and cannot certify form — roundness, flatness, squareness — because that would be the machine certifying its own geometry. The CMM measures in a controlled environment to finer resolution and is the authority for final acceptance. The two divide the work: probing screens the process in real time, the CMM judges the part. Used together they are stronger than either alone.

When is on-machine probing worth it? When your process produces the failures and setup hours probing removes: frequent changeovers, tight tolerances that need watching, multi-setup parts, drift-prone long runs, unattended or lights-out running, and large or variable parts that cannot easily reach a CMM. For simple, stable, loose-tolerance work in long runs, probing may never repay itself. The probe is a process tool — it pays only where the process uses it.

Bottom line

On-machine probing is the practice of letting the machine measure its own work while the part is still where it was cut — and it works when its nature is understood. The two probe families divide cleanly: the workpiece probe finds and checks the part, the tool setter sets and checks the tools. A touch probe is a switch, so everything it reports is interpreted through a calibration that must be maintained against a trusted artifact under the same conditions it measures in. The machine measures itself, which means probing certifies position and size but never form, and screens the process rather than judging the part. It operates before the cut to find the truth of the setup, during the cut to close the loop on drift and wear, and after the cut to screen what leaves the machine. It does not replace the CMM — it pairs with it, the rough screen beside the fine screen, cutting the sample rate and catching errors at the moment they can still be corrected. Understand the probe as a switch, maintain its calibration, respect what a machine measuring itself can and cannot know, and probing stops being an expensive gadget and becomes the difference between discovering an error on the third part and preventing it on the first.

This guide is part of the CNC Media guides library — the probing and on-machine-inspection reference of the measurement topic, deliberately free of prices and of any single probe maker’s catalogue to promote.