Neural networks for faster optical alignment

Abstract

A technique for providing neural network-coached alignment of optical systems is described in detail. The goal is to increase the speed of convergence to an aligned optical system. The neural network model is trained with the wavefront errors (WFEs) of thousands of randomly misaligned instances of the lens system that are modeled in Zemax OpticStudio®. The optical specialist measures the WFE of the misaligned system, then a neural network suggests specific adjustments to be made to the alignment fixturing (which adjuster, which direction, and by what amount). The technique has been developed into a MATLAB®-based tool called rapid optical system alignment with neural network assist that is shown to analytically and experimentally increase the speed of alignment. It is capable of achieving WFEs that are within a few percent of the nominally aligned condition. The ultimate goal is to deploy such a tool to enable an optical specialist to more quickly align challenging optical systems employing freeform or segmented surfaces. Analytical and experimental results for spherically symmetric systems are shown along with an outline of future work.