Design of an Online Measurement System for Combined Machine Tools
An online measurement system keeps the part on the machine and measures it there, so you skip the trip to the CMM between operations. This page covers the sensing chain, the thermal and datum errors that decide your real accuracy, and the part families where in-process gauging earns its keep. Written for engineers who have to pick a probe, a datum scheme and a sampling rate, not a brochure.
What an online measurement system actually measures
An online measurement system is a measuring loop that runs while the workpiece is still clamped on the machine. A touch probe or a non-contact sensor is loaded into the spindle or a dedicated holder, the machine moves the sensor to the feature, and the control records the trigger point in machine coordinates. The result is a position, not a size. Size only appears after you combine two or more positions with a known relationship.
That distinction decides your whole design. A probe with 1 μm repeatability still cannot tell you a bore is Ø50.02 mm unless the machine knows where the opposite wall is and how far the stylus traveled between the two triggers. So the design question is never "how accurate is the probe". It is "which positions do I trust, and what makes them drift".
Three error sources dominate on a combined machine tool. Thermal growth moves the spindle and the workpiece relative to each other over the warm-up cycle. Geometric error in the axes, including squareness and straightness, bends the path between two touch points. Probe and stylus deflection adds a small, repeatable bias that you can calibrate out. Everything else is small by comparison.
The engineering meaning is simple. An online measurement system is a relative instrument. It compares the part to a datum that the same machine established minutes earlier, in the same thermal state. If the thermal state changes between datum setting and measurement, the number you read is wrong even though every component worked perfectly.
Probe, stylus and signal chain choices
A touch-trigger probe is the default for prismatic parts. It gives a discrete trigger at contact, repeats well, and survives chips and coolant far better than optical sensors. For a 3-axis or 5-axis machine doing rough and finish in one setup, a spindle-mounted touch probe matched to the machine taper is usually the right call.
Stylus geometry sets what you can reach. A straight stylus with a Ø2 mm ruby ball reaches bores and slots that a Ø6 mm ball cannot enter. But a long stylus amplifies bending, so keep the effective length as short as the feature allows. For deep bores, expect to slow the touch feed and take more points per circle.
Non-contact sensors suit soft materials, thin walls and free-form surfaces where contact would deflect the part. The trade-off is sensitivity to surface finish, coolant mist and reflectivity. On machined aluminium with Ra 0.8–1.6 μm you get clean data. On a bead-blasted surface you do not.
The signal chain matters as much as the sensor. Hard-wired probe transmission gives the lowest latency, which matters when the probe must trigger at a controlled feed rate. Optical transmission is easier to retrofit but adds latency you must account for in the measuring macro. If the control cannot compensate for that delay, your readings shift with feed rate.
Why thermal drift sets your real accuracy
A machine that has run for 30 minutes is not the same machine it was at startup. The spindle grows, the ballscrews grow, and the bed can change shape as coolant and chips pile up. On a 4,000 mm machine the spindle-to-table relationship can move tens of micrometres over a warm-up cycle. That is larger than the tolerance you are trying to hold.
The standard fix is to set the workpiece datum after a warm-up period, then re-measure that datum before critical features. If the datum has drifted, shift your work offset instead of chasing the part. This is the single most effective habit in online measurement, and it costs one probe cycle.
Temperature compensation in the control helps, but only within its model. Compensation assumes a known coefficient of expansion for the casting and the screw. It cannot know that you opened the door for ten minutes or that a cold part was loaded onto a warm fixture. Sensors on the structure give you data; they do not give you a decision.
For tight work, measure the feature, then measure a reference feature that has not been cut since the last check. The difference between the two readings is thermal and geometric error, not part error. Subtract it before you touch the offset. Engineers who skip this step spend hours adjusting a machine that was already correct.
Datum strategy and sampling density
Every online measurement result is anchored to a datum. Pick a datum that is stable, accessible and machined in the same operation as the feature you are checking. A cast surface is a poor datum because its position varies from part to part. A bored hole or a machined face is a good one because the machine itself created it.
For a bore, four points give you center and diameter but not roundness. If roundness matters, take eight to twelve points and fit a circle. The fit tells you whether the error is a size offset, a center shift or an ovality problem, and each of those needs a different correction.
Sampling density is a cost decision. Each touch takes one to three seconds including approach and retract. On a part with twenty critical features, full coverage can add several minutes per cycle. Measure the features that drive fit and function, and measure them at enough points to separate the error modes you care about.
Sequence the measurement so the most drift-sensitive features are checked last, after the machine has reached thermal steady state. If a feature must be checked early, check it again at the end and compare. A single re-measurement is cheap insurance against shipping a part that was in tolerance at 9 a.m. and out of tolerance at 11 a.m.
When online measurement pays off and when it does not
Online measurement pays off when the cost of moving the part exceeds the cost of measuring it in place. Large parts are the classic case. A 4,000 mm frame that needs a crane and a re-fixturing setup to reach a CMM loses far more time in handling than it spends on probing. The same applies to parts that must stay in one setup to hold a relationship between features.
It also pays off on high-value parts where a late scrap decision is expensive. If a near-net forging or a titanium billet has hours of machining in it, finding a datum error after the finish pass is a disaster. Finding it after the first roughing pass costs a fraction of that.
It does not pay off on small, simple parts made in high volume. If a Ø30 mm bushing takes forty seconds to machine, adding a thirty-second probing cycle destroys the economics. For that work, a fixture that guarantees position and a sampling check at the end of the run is a better use of the machine.
It also does not pay off when the process is already stable and the tolerance is loose. Probing adds complexity, and complexity adds failure modes. If the last thousand parts held ±0.05 mm without a single probe cycle, adding one does not make them better. Use the data you already have from final inspection before you automate a check that never fails.
Matching the measurement method to the part
Pick the row that matches your part and production pattern.
| Part and pattern | Method | Why |
|---|---|---|
| Large frame, one setup, ±0.02 mm | In-process touch probe | Handling to a CMM costs more than probing |
| Small bushing, 50,000 per year | Fixture plus end-of-run check | Probe cycle would double the cycle time |
| Thin-wall housing, free-form | Non-contact sensor | Contact would deflect the wall |
| Deep bore, Ø40 mm × 300 mm | Long stylus, 12-point fit | Roundness needs a multi-point circle fit |
| Near-net titanium part | Probe after first roughing | Catch datum error before finish passes |
| Loose tolerance, stable process | Final inspection only | Probing adds failure modes for no gain |
The call: probe the part, not the machine
If handling dominates the cost, put a touch probe in the spindle and check the features that drive fit, with the datum re-set after warm-up. If the cycle time dominates, fixture the part hard and check it after the run. Do not probe to prove a process that already holds tolerance.
Questions engineers ask about online measurement
Can an online measurement system replace a CMM?
For process control, often yes. The probe tells you whether the machine is drifting and whether the part is where you expect it to be.
For final acceptance and certification, no. A CMM in a controlled environment with a calibrated scale gives you a traceable number. A machine probe gives you a number relative to that machine.
How many touch points do I need for a bore?
Four points give center and diameter. Eight to twelve points let you fit a circle and separate size error from center shift and ovality.
More points cost time. Match the count to the error mode you are trying to control, not to a habit.
Does the probe need to be re-calibrated?
Yes. Calibrate against a known ring gauge or setting sphere at the start of a run, and re-check after a crash or a stylus change.
Stylus deflection and ball wear both shift the trigger point over time, so calibration is a routine, not a one-time setup.
What coolant and chip conditions affect probing?
Chips on a datum face will shift your reading by their thickness. Blow off the feature before the probe approaches, and check that the air does not spray coolant onto the stylus.
On optical sensors, mist on the lens degrades the signal. Wipe or air-purge before each measurement cycle.
Can I probe a part while it is still hot from machining?
You can, but the reading includes thermal growth of the part. A hot Ø100 mm aluminium bore can read larger than it will be at 20 °C.
If size matters, either measure after a defined cooling period or measure the datum and the feature together and compare, so the thermal term cancels.
Does online measurement change the tolerance I can hold?
It changes what you can control, not what the machine can cut. A probe that resolves 1 μm does not make a machine that drifts 20 μm hold 5 μm.
Use it to find and correct drift. The floor on achievable tolerance is set by the machine, the fixturing and the thermal state, not by the sensor.
Send us the part and the tolerance
Tell us the feature list, the tolerance and the production volume. We will come back with a process plan, including where in-process measurement helps and where it just adds cycle time.
12-hour quote100% inspectionNo minimum order quantity
More from the shop floor
We publish setup notes, tooling trials and inspection data from the factory floor.
