Enhanced calibration for freeform surface misalignments in non-null interferometers by convolutional neural network

Area of Science:

• Optical metrology
• Interferometry
• Surface metrology

Background:

• Traditional surface calibration methods in interferometers, including sensitive matrix (SM) and deep neural network (DNN) approaches, often depend on Zernike coefficients.
• Non-circular interferograms arise in non-null freeform surface interferometry due to rotationally non-symmetric aberrations, even with circular surface apertures.
• Zernike polynomial-based methods exhibit inaccuracies in non-circular areas due to polynomial non-orthogonality.

Purpose of the Study:

• To address the limitations of Zernike coefficient-based calibration methods for freeform surfaces.
• To propose and validate a novel misalignment calibration method for interferometers using convolutional neural networks (CNNs).

Main Methods:

• A convolutional neural network (CNN) model was developed for direct interferogram analysis.
• The CNN was trained to estimate specific misalignments without relying on Zernike coefficients.
• Simulations and experimental setups were employed to test and validate the proposed CNN method.

Main Results:

• The proposed CNN-based method directly processes interferograms to estimate misalignments.
• The CNN approach overcomes the inaccuracies associated with Zernike polynomials in non-circular areas.
• Both simulations and experimental results demonstrate the high accuracy of the CNN-based calibration.

Conclusions:

• Convolutional neural networks offer a robust alternative for surface calibration in interferometers with non-circular interferograms.
• The direct analysis of interferograms by CNNs provides a more accurate estimation of misalignments in freeform surface metrology.
• This CNN-based method enhances the precision and reliability of surface calibration for complex optical components.