Abstract:
This thesis presents a novel approach to the alignment of Fabry-P´erot cavities using advanced
machine learning techniques. Precise cavity alignment is critical for applications ranging from
high-resolution spectroscopy and laser stabilization to gravitational wave detection. Traditional
manual alignment methods are often labor-intensive and prone to error, motivating the need for automated solutions. Also, we want to understand how effectively we can implement modern techniques in GW interferometer alignments. In this work, theoretical framework and simulations of a cavity setup was done to model the beam profiles present in the cavity. A convolutional neural network (CNN) model was developed to recognize the order of the mode images based on the simulation. Finally, this model was integrated with a Reinforcement Learning (RL) model to predict adjustments of the redirection mirrors to perfect the alignment. In the validation phase of the model, the model showed high precision in aligning the beam by adjusting the redirection mirrors and bringing the cavity mode to zero order TEM mode. This demonstrates that machine-learning based alignment systems can improve on the alignment efficiency and continuous realignment procedures. It will also reduce calibration time maintaining operational precision. This paves a way towards fully automated cavity alignment systems even in operational systems.
