Parallelism Check Laser Micrometer: How It Works and When to Use One

• 1
  What a parallelism check laser micrometer actually measures
• 2
  The optics behind the reading, and where error creeps in
• 3
  Single-head scan versus differential dual-head measurement
• 4
  Reading the number: what the report should tell you
• 5
  Where the laser method stops working
• 6
  Laser micrometer versus other parallelism checks

A parallelism check laser micrometer does not look at flatness, and it does not care about surface finish. It measures thickness at several points along a part and reports the spread. Two faces are parallel when that spread stays inside a band you set. If the top face is 20.004 mm above the bottom at one end and 20.011 mm at the other, the parallelism error is 0.007 mm over that span.

The beam is the ruler. An emitter projects a collimated fan of laser light across a gap, and a linear detector array on the far side reads where the part blocks the light. The shadow edges land on pixels, and the instrument converts edge position into a dimension. A scan takes a few milliseconds, so the head can sit in a production line and check every part instead of a sample.

Because nothing touches the part, thin walls, soft aluminium, polished faces and coated surfaces stay intact. A contact micrometer loaded on the same part can push 5–10 N into the surface and bend it. On a 1 mm aluminium wall that squeeze alone can move the reading by more than the tolerance you are trying to hold.

One limit is worth stating early. The shadow tells you the outer envelope between two edges. It cannot see a concave face that dips in the middle unless the dip reaches the outline. Internal parallel faces, blind slot floors and bores are outside what a single through-beam head can measure.

• 1
  What it gives youThickness spread across a known scan path, expressed in mm or μm per unit length.
• 2
  What it ignoresFlatness, waviness below the scan resolution, subsurface defects and internal features.
• 3
  Best fitThrough-features and outer faces where both sides of the part are reachable by the beam.

A laser micrometer works on edge detection. The detector is a linear array of photodiodes, often 1,024 or more elements on a 30 mm window. When the part interrupts the beam, the boundary between lit and unlit pixels marks the edge. The instrument applies a sub-pixel interpolation to place that boundary, which is how a 30 mm window resolves down to a fraction of a micrometre.

Edge quality sets the floor. A burr, a chamfer or a rolled edge shifts the apparent boundary. So does a shiny surface that scatters light back toward the emitter. Most heads ship with a diffuse or retro-reflective mode for exactly this reason. If your part comes off a polished anodize line, expect a small offset until you qualify the setup with a known master.

Alignment matters more than resolution. If the beam is not normal to the part axis, the measured thickness is the true thickness divided by the cosine of the tilt. At 1° that is a 0.015% error, which is harmless. At 5° it is 0.38%, and on a 20 mm part that is 0.076 mm of phantom error. Fixture the part, not the operator's hand.

Temperature is the slow error. Aluminium grows about 23 μm per metre per °C. A 20 mm aluminium part that warms 5 °C between the machine and the gauge reads 2.3 μm thicker. That is inside a ±0.005 mm tolerance, so a shop that checks parts on a warm bench and ships them to a 20 °C inspection room will see disagreements.

• 1
  Beam normalityKeep tilt under 1° or correct the reading by the cosine factor.
• 2
  Edge conditionDeburr and break edges to a consistent 0.1–0.2 mm chamfer.
• 3
  Thermal soakLet parts sit at 20 °C ±1 °C for 30 minutes before the final scan.
• 4
  CleanlinessChips and coolant film change edge position by 1–5 μm.

One head measures one edge pair at a time. Move the part or the head along the length, log thickness at each station, and subtract the minimum from the maximum. That is the classic parallelism error: max thickness minus min thickness over the measured area. It is simple, and it is enough for most flat plates and rails.

Two heads facing each other do something better. One measures the top-to-bottom envelope while the other measures a reference edge, and the system subtracts the two signals. Common-mode errors cancel. Vibration on the fixture, a drift in ambient temperature, a slow change in laser power — all appear in both channels and drop out of the difference.

The payoff is repeatability. A single-head setup in a busy shop might repeat to 1–2 μm over a shift. A differential pair can hold sub-micron repeatability because the reference channel moves with the noise. The trade is cost and setup time: two heads must be aligned to each other, and that alignment itself needs a master.

Scanning along the length adds a third option. Instead of point checks, the head travels the part and builds a thickness map. The map shows wedge, convexity, concavity and twist separately. A wedge part is thick at one end. A twisted part is thick at opposite corners. Those are different machining faults and they need different corrections.

• 1
  Single headLowest cost, good for 2–10 μm parallelism bands on stable parts.
• 2
  Dual headBest repeatability, needed when the band is under 2 μm.
• 3
  Scanning mapSeparates wedge, bow and twist; useful for first-article reports.

A parallelism result without a stated span is meaningless. 0.008 mm over 20 mm is a different part from 0.008 mm over 200 mm. Ask for the scan length, the number of stations and the datum face. If the report says only "parallel within 0.01 mm", you cannot tell whether the gauge scanned the full face or one short track.

Station spacing matters too. A part can pass a two-point check and fail a ten-point scan, because roughness or a local low spot sits between the two points you measured. For a 100 mm face, 10 mm spacing gives 11 stations and catches most bow. For a 400 mm rail, 25 mm spacing is a reasonable starting point.

Watch for the difference between parallelism and thickness variation. They use the same data but answer different questions. A part with a uniform 0.05 mm taper is perfectly parallel in the sense that both faces stay the same distance apart — it is just the wrong size. Parallelism is about the spread, not the nominal.

Finally, confirm the gauge was qualified on the day. A master disc or gauge block check before and after the run takes two minutes and catches drift. Shops that skip it tend to discover the drift in the customer's incoming inspection instead.

• 1
  Demand the spanParallelism without a length is not a specification.
• 2
  Check station countTwo points hide bow; ten points usually do not.
• 3
  Separate size from formTaper and parallelism are different failures.
• 4
  Qualify dailyA master check before and after the run costs two minutes.

Small features are a hard stop. Most heads need a clear beam path around the part, and the window is typically 5–30 mm. A 3 mm wide rib between two walls cannot be scanned without the walls clipping the beam. For those, a vision system or a CMM with a small stylus is the right tool.

Internal parallel faces are another stop. A slot floor and its mating top face inside a housing cannot be reached by a through-beam. You can measure the outside envelope and infer, but inference is not inspection. If the drawing calls out parallelism on an internal face, plan for a tactile or optical probe.

Very rough or heavily textured surfaces scatter the beam and blur the edge. As-machined faces at Ra 1.6–3.2 μm scan fine. Sandblasted or thermal-sprayed surfaces often need a different technique, or a polished witness area added to the drawing for measurement only.

And the method says nothing about stiffness, residual stress or how the part behaves when bolted down. A plate can pass a free-state parallelism check and twist when four bolts pull it flat. If the function depends on the assembled state, the check has to happen in the assembled state.

• 1
  Feature too smallBeam path narrower than the head window cannot be scanned.
• 2
  Feature internalBlind slots and bores need a probe, not a through-beam.
• 3
  Surface too roughHeavy texture scatters the edge; add a witness area.
• 4
  Free state onlyAssembled parallelism needs an assembled check.