Automatic Classification of Post Mortem Data for Reduced Beam Down Time

Time spent recovering from faults that result in a rapid loss of stored current (a total beam loss event) can be costly to the Australian Synchrotron facility and its researchers. The identification of a fault leading to total beam loss is assisted by a large variety of investigative tools for specific tasks, but they do not often give a thorough overview of all systems required to store beam. Post mortem data uniquely provides insight into how the beam was behaving at the specific time the dump occurred. With machine learning, we find that we can automatically and rapidly identify many types of total beam loss events by learning about the unique characteristics in the post mortem files.