Back to Portfolio

Machine Learning for Fabry–Pérot Cavity Alignment

Developed a reinforcement learning framework to auto-align a Gaussian beam in zero-order TEM mode into a Fabry-Pérot cavity with sub-20μm accuracy.

Machine LearningAstrophysicsInstrumentation

Overview

For my MS thesis, I developed a machine learning model capable of auto-aligning a Gaussian beam into a Fabry-Pérot cavity using steering mirrors.

Method

  • Created a laboratory setup with a Fabry-Pérot cavity and two steering mirrors with two degrees of freedom each.
  • Performed mode matching to remove Laguerre-Gaussian mode content up to 98%.
  • Developed a reinforcement learning model coupled with a CNN to analyze mode images in the cavity with 100% accuracy.
  • Achieved sub-20μm alignment error.

Custom Designs

  • Designed and 3D-printed NEMA motor mounts for the mirrors using FreeCAD.
  • Constructed a beam profiler achieving an error margin of ±15 microns.

Impact

The integrated system demonstrated practical application of reinforcement learning in automated, high-precision optical setups—potentially benefiting systems like LIGO and VIRGO.