Automatic Classification of Pressure Readings using Analytical and Machine Learning techniques

Along the Large Hadron Collider (LHC) over 3000 Vacuum Gauges (VG) help ensure the integrity of the Ultra High Vacuum (UHV) is maintained. In a minority of cases the pressure readings from these gauges exhibit anomalous behaviour where pressure does not gradually decrease as expected. Through pre-processing and supervised learning, this study obtains a 92% accuracy in classifying typical and anomalous behaviour using a K-Neighbours algorithm. Consequently, the number of gauges needing to be diagnosed is constrained and faulty gauges or potential sources of power loss can be identified.