Can a neural network determine the beam offset in an X-ray polarimeter?

Sammanfattning

By measuring the polarization of X-ray emission from distant stellar objects such as neutron stars and stellar-mass black holes, information about their geometry and properties can be discerned. XL-Calibur is a balloon borne polarimetric mission that utilizes Compton scattering to determine the polarization of the emitted X-rays. The incoming photons are focused into the polarimeter by an X-ray mirror which is all held in alignment by an optical bench (truss). Factors during flight such as uneven heat exposure or gravitational sag can cause a slight offset in the alignment of incoming photon beam introducing a systematic error in the reconstructed polarization. The aim of this thesis is to analyze if an artificial neural network can use the data collected of photon hits in the polarimeter to correctly predict this offset. The neural network was trained and tested using data simulated to have the same point spread function (PSF) as the X-ray mirror. The model was then applied to the data collected from the XL-Calibur flight. The model performed exceedingly well on simulated data but further studies are required to determine the validity of the network’s predictions on flight data.