[2Ip01-07] Vacuum Science Technology (VST)

[2Ip05] Modeling of pressure behaviors for a pressure anomaly detection program utilizing machine learning in the SuperKEKB accelerator

*Yusuke Suetsugu¹ (1. High Energy Accelerator Research Organization (KEK))

An anomaly pressure detection program, employing machine learning in the SuperKEKB accelerator, is being proposed and is currently under development. Regression curves, describing pressure behavior during a normal state (reference data) as a function of beam current or time, are derived using appropriate models. By utilizing the ratio of the root mean square error (RMSE) of the data to be evaluated (check data) and others relevant factors as input parameters, a two-layer feedforward neural network (FNN) is constructed. This network classifies the check data into two categories: "normal" and "abnormal". A pivotal element in the development of the program involves crafting suitable models for pressure behavior in order to generate precise regression curves. This report presents a pressure anomaly detection program employing machine learning, with a focus on a straightforward yet rational approach to modeling pressure behaviors for deriving regression curves.