Accelerators and their ghosts

The issue of particle accelerator reliability is a problem that currently is not fully defined, understood nor addressed. Conventional approaches to reliability (e.g., RBDs) struggle due to a lack of data about specific component/system reliability and failure.

There is a large body of beam current data retrievable from operating accelerators that contains detailed information about the accelerator behaviour, both before and after a
machine trip has occurred.

Analysing this data could provide insight and help develop a new approach to address accelerator reliability. In this paper, we propose a data-driven approach to detecting emergent behaviour in particle accelerators. Instead of attempting to identify every possible failure of a machine
we propose an alternative approach based around a change in perspective, to knowing the normal default operational behaviour of a machine. Taking action when a “ghost in the machine” emerges that causes accelerator wide aberrant changes to normal machine behaviour.