Modern gravitational wave detectors such as LIGO are the most sensitive measurement devices ever built. They rely on nested optical cavities, with each mirror suspended to limit seismic motion, leaving mirrors free to move, but positions must still be precisely sensed and controlled, a rich problem with many interacting degrees of freedom. Initial alignment is especially hard: after earthquakes or maintenance, mirrors can become severely misaligned, and manually recovering alignment is slow and skill-dependent.

We want to apply machine learning to this problem, finding optimal solutions faster and more consistently than by hand. Building on recent Centre work combining a convolutional neural network (CNN) with a genetic algorithm (GA) to align a table-top cavity, this project extends that approach. A camera monitors transmitted light; the CNN classifies the mode in real time, and the GA uses this and the power as a reward signal to search mirror positions maximising fundamental-mode power.

This already recovers alignment from a random start within a handful of generations, matching or exceeding manual quality, using just a single camera rather than photodiode sensing. The student will extend this by testing against a simulation before demonstrating on a table-top experiment, aiming toward the suspended cavities in real detectors.

The student will build skills in machine learning (CNNs, genetic algorithms, RL-style optimisation), optics, cavity alignment, and control systems — ideal for a student interested in AI/ML and experimental physics.

Get in touch to discuss research topics — see the link below, or visit the Centre for Gravitational Astrophysics.

Some knowledge of python or another imperative programming language, image processing experience would be useful.