Does the laser interferometer work? How the reading is actually built

• 1
  How a laser interferometer turns light into length
• 2
  Why air is part of the measuring instrument
• 3
  The four setup mistakes that fake a good result
• 4
  What the linear error plot tells you about the machine
• 5
  When the laser interferometer is the wrong tool
• 6
  A repeatable linear measurement run
• 7
  Which measurement method fits which job

A laser interferometer splits one beam into two. One part travels to a fixed reflector, the other to a reflector mounted on the moving axis. When the two beams come back together they interfere, and the detector sees a bright or dark band depending on the phase difference. Move the axis by half a wavelength and the pattern flips once.

That is the whole trick. The instrument does not know where the axis is. It counts how many times the pattern flips and multiplies by half the laser wavelength. On a helium-neon source the wavelength sits near 633 nm, so one count is roughly 0.316 µm of movement.

The count is why the method works at all. Instead of measuring a long distance with one comparison, the system measures a very short distance very many times. A 1,000 mm axis travel becomes millions of counted fringes, and each one is a clean electrical pulse.

The same idea gives the technique its weak point. Anything that changes the optical path length without moving the axis will be counted as motion. Air, heat, and vibration all do exactly that.

• 1
  Half wavelength per countOne fringe equals about 0.316 µm on a 633 nm source.
• 2
  Counts, not distancesThe display is an accumulated fringe count, not a physical measurement.
• 3
  Both beams matterAnything altering either path length enters the result.

Light slows down in air. The amount depends on air density, which depends on temperature, pressure and humidity. Change the temperature by 1 °C across a 1 m beam and the wavelength in air shifts by roughly 0.9 ppm, which is 0.9 µm over that metre. That is larger than the tolerance of many parts we machine.

So the sensor package is not optional. A typical setup carries an air temperature sensor, a barometric pressure sensor and a humidity sensor, and the controller applies the Edlén equation to correct the wavelength continuously. Skip that correction and you are measuring the weather as much as the axis.

Air turbulence matters more than average conditions. A beam passing near a spindle motor, a servo drive or a warm ball screw sees a moving patch of hot air. The reading drifts and returns, which looks like axis noise but is not. This is why the beam should run as far as possible from heat sources.

A simple test tells you whether the environment is stable enough. Park the axis and log the reading for a few minutes. If the display wanders more than a tenth of your target uncertainty, fix the environment before measuring motion.

• 1
  1 °C ≈ 0.9 ppmAbout 0.9 µm per metre of beam on a 633 nm source.
• 2
  Log a stationary axisDrift at standstill is the noise floor of the whole setup.
• 3
  Keep the beam off hot surfacesSpindle motors and drives create local air gradients.

Dead-path error is the first and the most common. It is the distance between the beam splitter and the moving reflector at the moment you zero the display. If that gap is 100 mm and you forget to enter it, every reading is offset by roughly the expansion of that 100 mm of air, and the offset scales with temperature.

Cosine error comes next. If the beam is not parallel to the axis of travel, the measured length is the projection, not the true travel. A 1 mm lateral offset over 1 m of travel gives about 0.5 µm of under-reading. That sounds small until you are chasing a 2 µm spec.

Abbé error is a geometry problem, not an optics problem. If the laser beam sits 200 mm above the ball screw and the machine pitches by 20 µm/m, the beam sees a different displacement than the tool point does. The instrument is right; it is measuring the wrong line.

Finally there is the material question. The wavelength in vacuum is fixed, but the workpiece and the machine frame expand with temperature. A steel part grows about 11.7 µm per metre per degree Celsius. At 1 m and a 5 °C shop swing, that is 58 µm of real movement that no optical correction can remove.

• 1
  Enter the dead pathMeasure splitter-to-reflector distance and type it into the controller.
• 2
  Align to the axis, not the tableUse the machine travel direction as the reference line.
• 3
  Mind the Abbé offsetKeep the beam close to the tool point line where possible.
• 4
  Let parts soakThermal equilibrium takes longer than most people expect.

A linear measurement run gives you a plot of commanded position against measured error. On a healthy machine that plot is close to a straight line with a small slope. The slope is the screw pitch error, and it is the part a controller can compensate with a simple linear correction.

Curvature is more interesting. A bowed plot usually means the ball screw is not uniform, the screw is running hot in the middle of travel, or the axis is being pushed by a guideway that is not straight. Straightening the compensation table will hide it, not fix it.

Reversal spikes are the third signature. Run the axis in one direction, stop, then reverse. The plot jumps by the backlash plus the elastic wind-up of the drive train. Typical values on a mid-size vertical machining centre run from a few micrometres to tens of micrometres depending on preload and wear.

Repeat the run at least twice in the same direction and once in reverse. A single pass tells you the error. Two passes tell you whether the error is stable enough to compensate. If the two runs disagree, the problem is thermal or mechanical, and compensation will chase the machine around.

• 1
  Straight slopeUniform pitch error, the easy case for linear compensation.
• 2
  Bowed curveNon-uniform screw heating or guideway geometry.
• 3
  Reversal stepBacklash plus drive-train wind-up.
• 4
  Repeat before compensatingUnstable error cannot be corrected with a static table.

A laser interferometer measures along a line. If you need to know how a machine behaves while cutting a curved path at feed, the interferometer will not tell you. It cannot see servo lag under load, tool deflection, or the interaction between two axes moving at once. A ballbar test or a test cut covers that ground.

It also struggles in a working shop. Fog coolant, grinding dust and airborne oil all scatter the beam or coat the optics. A machine that is mid-production is often a bad place to set up a 10 m optical path. Many shops measure during a planned stop, with coolant off and the enclosure closed.

Very long axes bring their own problem. Over 4 m the beam is walking through a lot of air, and the environmental correction becomes the dominant uncertainty. At that length a scale-based verification or a multi-target tracker is usually more practical.

Finally, remember what the number means. A linear error of 8 µm over 1,000 mm is a statement about a specific beam path at a specific temperature. It is not a statement about the parts the machine will produce tomorrow morning. Use it as a maintenance input, not as a final part inspection.

• 1
  No contouring dataInterferometry is a single-axis, line-of-sight method.
• 2
  Coolant and dustOptics need a clean path and clean glass.
• 3
  Long travel, big air pathEnvironmental uncertainty grows with beam length.
• 4
  It is not part inspectionMachine error and part error are different quantities.