Measuring the ELT M2 mirror: a custom interferometric test bench

The secondary mirror (M2) of the Extremely Large Telescope (ELT) is the largest convex precision mirror ever made. Polishing it requires measuring it regularly with extreme accuracy. At Safran Reosc, we developed a dedicated interferometric test bench for this.

My contribution: I designed the control system and software architecture of the bench. I then supervised their development, and the system validation.

Why a custom bench

• A convex mirror is hard to measure: no instrument can be placed at its centre of curvature.
• The usual solution is a test matrix: a lens whose concave face has exactly the shape of the mirror and serves as the reference.
• But nobody can make an optical-quality lens as large as the mirror. The matrix only covers part of it, so the mirror is measured in sub-apertures, then stitched together to rebuild the full map.

The bench architecture

• The mirror sits on its active support, described in this article, on a turntable that rotates it.
• The test matrix is placed very close to the mirror. This very short cavity limits air disturbances.
• The interferometer stands at the top of a tower above the mirror. It makes the beam reflected by the matrix interfere with the beam reflected by the mirror.
• A multi-axis motorised alignment drives the matrix with high-resolution actuators. It relies on a position sensor (PSD) lit by a laser mounted on the mirror, and on LVDT sensors that measure the cavity thickness.
• The bench is fully automated: a complete measurement, stitching included, takes a few hours.

An exceptional interferometer and test matrix

• The interferometer, custom designed, acquires its images very fast, which makes it robust to vibrations. Its low-coherence source removes parasitic fringes.
• It is extremely repeatable, and sees the whole spectrum of mirror defects in a single image.
• The test matrix in ZERODUR® is one of the largest and most difficult lenses ever made: its concave face is aspheric and its convex face is a freeform surface.
• It took several years to make. It is now almost perfect, and its residual defects are known and corrected in the measurements.

Alignment and stitching

• The stitching algorithm adjusts each sub-aperture to minimise the differences in the overlap areas.
• Simulations show that it exactly recovers a deliberate decentre of the matrix, without adding any error to the final map.
• On a real measurement, it corrects the small alignment differences between sub-apertures: the matrix trajectory becomes the expected one again.

Validating the bench

We built a statistical error budget from a very large number of contributors (mechanics, optics, environment, data processing). We then carried out the validation plan on the M2 mirror itself.

• Sub-aperture repeatability: hundreds of successive measurements stay clearly below the set criterion.
• Full measurement repeatability: measurements repeated over several days stay well below the requirements, with no difference between short and long term.
• Reproducibility: after removing the mirror, rotating its support and putting the mirror back, the measurements agree within the criteria.
• Cross-check with a coordinate measuring machine: both methods agree far better than required.

Where the mirror stands

• After several polishing iterations, all low-frequency requirements are met.
• The residual surface error is already clearly better than the specification, and polishing now targets the last high-frequency defects.
• The sub-apertures match very well, a sign of the stitching quality.

Key takeaways

• All the performance required from the bench is met, often with a wide margin.
• My takeaway as the architect: a fully automated bench makes it possible to repeat measurements and prove its accuracy.
• The mirror polishing converges faster than expected.

Further reading

R. Mercier-Ythier, H. Monin, G. Foucaud, C. Beudy, N. Lefeuvre et al., Development of an interferometric test bench for the manufacturing of the ELT secondary mirror, Proc. SPIE 13021-17 (2024), SPIE · DOI