BNL, Upton, New York, USA
FRIB, East Lansing, Michigan, USA
The Relativistic Heavy Ion Collider (RHIC) Cryogenic Refrigerator System is the cryogenic heart that allows RHIC superconducting magnets to operate. Parts of the refrigerator are two stages of compression composed of ten first and five second-stage compressors. Compressors are critical for operations. When a compressor faults, it can impact RHIC beam operations if a spare compressor is not brought online as soon as possible. The potential of applying machine learning to detect compressor problems before a fault occurs would greatly enhance Cryo operations, allowing an operator to switch to a spare compressor before a running compressor fails, minimizing impacts on RHIC operations. In this work, various data analysis results on historical compressor data are presented. It demonstrates an autoencoder-based method, which can catch early signs of compressor trips so that advance notices can be sent for the operators to take action.
BNL, Upton, New York, USA
The successful use of machine learning (ML) in particle accelerators has greatly expanded in recent years; however, the realities of operations often mean very limited machine availability for ML development, impeding its progress in many cases. This paper presents a framework for exploiting physics-based simulations, coupled with real machine data structure, to facilitate the investigation and implementation of reinforcement learning (RL) algorithms, using the longitudinal bunch-merge process in the Booster and Alternating Gradient Synchrotron (AGS) at Brookhaven National Laboratory (BNL) as examples. Here, an initial fake wall current monitor (WCM) signal is fed through a noisy physics-based model simulating the behavior of bunches in the accelerator under given RF parameters and external perturbations between WCM samples; the resulting output becomes the input for the RL algorithm and subsequent pass through the simulated ring, whose RF parameters have been modified by the RL algorithm. This process continues until an optimal policy for the RF bunch merge gymnastics has been learned for injecting bunches with the required intensity and emittance into the Relativistic Heavy Ion Collider (RHIC), according to the physics model. Robustness of the RL algorithm can be evaluated by introducing other drifts and noisy scenarios before the algorithm is deployed and final optimization occurs in the field.
