In-line measurement
In-line measurement — In-line measurement is a measurement taken on the moving assembly without removing the part, so the sensor must deliver a decision inside the takt time and cannot use the settling time, re-fixturing or multiple readings that a laboratory measurement relies on.
Gauge repeatability and reproducibility
Gauge repeatability and reproducibility — Gauge repeatability and reproducibility splits the spread of a measurement system into the variation one operator or station produces on the same part and the variation added by different operators, stations or times, and the whole spread is what must stay small against the tolerance band.
| Station | What the sensor decides | Dominant error source | Effective countermeasure |
|---|---|---|---|
| Weld check | Seam gap and presence of material | Spatter and arc radiation landing on the sensor | Air purge, viewing angle off the weld axis, narrow band filter |
| Bolting | Presence and height of a fastener before the head is seated | Surface reflection from a plated head | Diffuse illumination, incidence far from the mirror angle |
| Fit and gap | Flush between two panels | Thermal growth of the part, not the sensor | Measure at a stated temperature, control fixturing repeat |
| Pick and place | Pose of a loose part in a tray | Ambient light changes over the shift | Banded illumination and a trigger the robot owns |
| Tolerance band to resolve | 0.20 mm |
|---|---|
| Allowable share of band used by the sensor | 10 percent |
| Working distance available | 180 mm |
| Cycle available per decision | 3.5 s |
| Part material | anodized aluminium |
| Ambient condition | daylight through roof lights |
- Growth of a 600 mm aluminium bracket over a 15 K shift
- 0.14mm
- Source: Computed from a nominal expansion coefficient of 2.3e-6 per kelvin for the example length. Illustrative sample dataset, not a measured value.
Is a 0.01 mm resolution sensor enough to control a 0.2 mm band?
Resolution is not the constraint, and specifying on it is how plants end up with an expensive gauge that still cannot judge the part. The useful question is the width of the whole measurement distribution, including fixture repeat, part temperature and cable or bracket compliance; if that distribution occupies most of the tolerance band, the readings will disagree with the coordinate measuring machine no matter how many decimal places the sensor displays.
Why does a laser sensor read a shot-blasted surface as rough?
Because the surface is a field of tiny facets, each returning light at its own angle, so the speckle pattern changes with position and the sensor averages a jagged return rather than a flat one. The remedies are to increase the spot so more facets contribute and average out, to choose a shorter wavelength or a confocal principle that rejects scattered light, or to accept the texture in the setup and measure a relative feature instead of an absolute height.
Does welding current interfere with an analogue in-line sensor?
It can, through two separate paths that need different fixes. The magnetic field of the weld current induces a voltage in any loop the signal cable forms with the machine, which is reduced by twisting the pair and routing it against the frame, while a difference in earth potential between the welder and the control cabinet drives current through the screen, which is handled by bonding the screen at one end and by choosing a sensor whose input is not referenced to the same earth.
All figures on this page, including the illustrative expansion example in the stat block and the task parameters in the list, are example values for a demonstration site. Replace them with your own measured data before publishing, and do not use this page as a specification.
Start from the tolerance band, not the sensor
An in-line sensor is specified backwards in most plants: somebody selects a device with a resolution that looks impressive, then discovers the station still rejects good parts. The correct order starts with the band to be judged, and how much of it the measurement system is allowed to consume. If the band is 0.20 mm and the measurement system may use 10 percent of it, the total distribution — not the displayed last digit — has to sit inside 0.02 mm, which is a demand on the fixture and the part temperature as much as on the optics.
Two consequences follow. Comparison against a laboratory instrument is a comparison of distributions, not of single readings, so a gauge that agrees with the coordinate machine on one part has proved nothing. And a station whose rejection rate is driven by the measurement system rather than the process will show up as parts that fail on the line and pass when re-measured offline, which is the classic signature to look for before buying hardware.
Thermal growth is bigger than the tolerance
Consider a bracket 600 mm long that arrives from a wash line warmer than the shop floor, or that has been welded locally. Aluminium grows at roughly 2.3e-6 per kelvin of nominal coefficient, so a 15 K difference is about 0.14 mm at full length — a large share of any fit-and-gap band. Steel grows at about half that rate, so a bimodal assembly changes its relative geometry as it equalises, and a sensor that is perfectly repeatable will still report a different number in the morning than in the afternoon.
The engineering response is not a better sensor. It is to state the measurement temperature in the work instruction, to place the measurement after the part has equalised or at a fixed dwell, and to keep the datum the sensor references on the same thermal path as the feature it measures. Where that is impossible, measure the differential feature — flush, gap, step — between two nearby surfaces, because a common-mode length change cancels in a differential reading and survives in an absolute one.
Choosing the principle at each station
Weld and weld-check stations fight light that is not yours: arc radiation is intense in the visible and near ultraviolet, and spatter is molten and moves. A triangulation sensor mounted in the spatter path will read spatter as part geometry until it is cleaned. Air purge, a consumable protective slide, and a viewing direction that keeps the weld axis out of the specular reflection path are the effective measures; a narrow band filter matched to the illumination laser removes most of the rest.
Bolting stations are a reflection problem rather than a distance problem: plated heads are close to mirrors, and a beam returned at the mirror angle saturates the receiver while the neighbouring channel sees nothing. Off-set the incidence, diffuse the source, or measure the feature that is diffuse — the shank, the hole edge, the marking on the washer.
Fit-and-gap is a fixture problem with an optical label. The panels are held by suction cups and clamps whose repeat accuracy is often worse than the sensor’s, so the honest first improvement is a repeatability study of the fixture and hard stops rather than more decimal places.
What has to be true before the line runs
Three items decide whether the installation behaves, and all three are cheap before commissioning and expensive after. First, illumination must be controlled by the machine rather than the room: a hood, a banded filter and a trigger the robot owns remove daylight, forklift headlights and the robot’s own welding flashes from the error budget. Second, the signal path must be planned as a loop: keep analogue runs short and twisted, keep them out of the same duct as welder and drive cables, bond screens at one end only, and give the sensor its own reference rather than borrowing the welder’s. Third, the maintenance window must exist on paper: a lens with a purge orifice, a mount with one adjustment that is locked and witness-marked, and a check part whose value is measured offline on a fixed cadence, so drift is seen as drift and not as a mysterious process change.