Optical alignment procedure utilizing neural networks combined with Shack-Hartmann wavefront sensor

Adil F. Z., Konukseven E. I., Balkan T., Adila O. F.

OPTICAL ENGINEERING, cilt.56, sa.5, 2017 (SCI-Expanded) identifier identifier

  • Yayın Türü: Makale / Tam Makale
  • Cilt numarası: 56 Sayı: 5
  • Basım Tarihi: 2017
  • Doi Numarası: 10.1117/1.oe.56.5.051402
  • Dergi Adı: OPTICAL ENGINEERING
  • Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus
  • Anahtar Kelimeler: wavefront, Shack-Hartmann sensor, optical system alignment, Zernike polynomials
  • Orta Doğu Teknik Üniversitesi Adresli: Evet

Özet

In the design of pilot helmets with night vision capability, to not limit or block the sight of the pilot, a transparent visor is used. The reflected image from the coated part of the visor must coincide with the physical human sight image seen through the nonreflecting regions of the visor. This makes the alignment of the visor halves critical. In essence, this is an alignment problem of two optical parts that are assembled together during the manufacturing process. Shack-Hartmann wavefront sensor is commonly used for the determination of the misalignments through wavefront measurements, which are quantified in terms of the Zernike polynomials. Although the Zernike polynomials provide very useful feedback about the misalignments, the corrective actions are basically ad hoc. This stems from the fact that there exists no easy inverse relation between the misalignment measurements and the physical causes of the misalignments. This study aims to construct this inverse relation by making use of the expressive power of the neural networks in such complex relations. For this purpose, a neural network is designed and trained in MATLAB (R) regarding which types of misalignments result in which wavefront measurements, quantitatively given by Zernike polynomials. This way, manual and iterative alignment processes relying on trial and error will be replaced by the trained guesses of a neural network, so the alignment process is reduced to applying the counter actions based on the misalignment causes. Such a training requires data containing misalignment and measurement sets in fine detail, which is hard to obtain manually on a physical setup. For that reason, the optical setup is completely modeled in Zemax (R) software, and Zernike polynomials are generated for misalignments applied in small steps. The performance of the neural network is experimented and found promising in the actual physical setup. (C) 2017 Society of PhotoOptical Instrumentation Engineers