Force controlled mid-spatial frequency error smoothing using machine learning

Abstract

Freeform optics have shown the ability to drastically increase optical design options, providing new solutions for space telescope and optics applications. Freeform optics manufacturing strives to provide designers this flexibility, but the requisite subaperture generation and polishing techniques often result in mid-spatial frequency (MSF) errors. Because these errors negatively impact final performance, they must be removed. However, mitigation options are costly and timeconsuming. The HERMES (High End Robotic MSF Elimination System) is a robotic platform that uses a machine learning algorithm and deflectometry data to smooth optics. This project has shown preliminary success, and began to forecast increases in form error to help balance smoothing time predictions. Other advancements are explored by the current work. First, the algorithm has been upgraded to include a force-controlled option that utilizes form data (from interferometry) to scale applied force during the smoothing process that is mapped to deviation data. This project also integrated an orthogonal (or raster) pattern of smoothing found in early human smoothing recordings. The various conditions were compared, and result metrics included the peak power spectral density (PSD) values of the frequency of interest as well as irregularity (form) error. All conditions decreased peak PSD values reliabilty, but irregularity data became more difficult to interpret. Findings and further work are discussed.