CERN, Geneva, Switzerland
A risk matrix is a common tool used in risk assessment, defining risk levels with respect to the severity and probability of the occurrence of an undesired event. Risk levels can then be used for different purposes, e.g. defining subsystem reliability or personnel safety requirements. Over the history of the Large Hadron Collider (LHC), several risk matrices have been defined to guide system design. Initially, these were focused on machine protection systems, more recently these have also been used to prioritize consolidation activities. A new data-driven development of risk matrices for CERN’s accelerators is presented in this paper, based on data collected in the CERN Accelerator Fault Tracker (AFT). The data-driven approach improves the granularity of the assessment, and limits uncertainty in the risk estimation, as it is based on operational experience. In this paper the authors introduce the mathematical framework, based on operational failure data, and present the resulting risk matrix for LHC.
TUB, Beijing, People’s Republic of China
NINT, Shannxi, People’s Republic of China
Fermilab, Batavia, Illinois, USA
JLab, Newport News, Virginia, USA
We describe the offline machine learning (ML) development for an effort to precisely regulate the Gradient Magnet Power Supply (GMPS) at the Fermilab Booster accelerator complex via a Field-Programmable Gate Array (FPGA). As part of this effort, we created a digital twin of the Booster-GMPS control system by training a Long Short-Term Memory (LSTM) to capture its full dynamics. We outline the path we took to carefully validate our digital twin before deploying it as a reinforcement learning (RL) environment. Additionally, we demonstrate the use of a Deep Q-Network (DQN) policy model with the capability to regulate the GMPS against realistic time-varying perturbations.
ORNL RAD, Oak Ridge, Tennessee, USA
ORNL, Oak Ridge, Tennessee, USA
The Spallation Neutron Source (SNS) is an accelerator based pulsed neutron source based on a 1 GeV pulsed proton Superconducting Radio Frequency (SRF) linear accelerator (linac). Since beginning high power beam operation in 2006 correlations have been found linking abrupt beam loss events to SRF cavity instabilities. With the planned upgrades to double the beam power we expect increased rates of degradation and the importance of minimizing these beam loss events will become ever more important. To further limit degradation, we are developing machine learning approaches to monitor the beam and to detect, predict and prevent beam loss events. Initial research has shown that precursors to beam loss events are detectable. The initial steps are to use ML-based classification to recognize anomalies and to use Long Short-Term Memory (LSTM) autoencoders to predict beam loss. In this paper, we describe recent progress in applying machine learning for recognizing anomalies and predicting beam loss and present initial results of our research using acquired data from different diagnostics and the Machine Protection System (MPS).
ORNL RAD, Oak Ridge, Tennessee, USA
ORNL, Oak Ridge, Tennessee, USA
JLab, Newport News, Virginia, USA
PNNL, Richland, Washington, USA
Beam availability has increased at the SNS, however, the targeted availability is greater than 95 %, while the SNS has failed to meet lower targets in the past. The HVCM used to power the linac klystrons have been one source of lost beam time and was chosen to explore using AI/ML techniques to improve reliability. Among the possibilities being explored are automating the tuning of HVCMs and predicting component failures such as capacitor aging, rectifier assemblies containing hundreds of diodes, and insulating oil degradation. The methodology pursued includes data cleaning, de-noising, post-analysis data labeling, and machine learning model development. We explore using Long Short-Term Memory and autoencoders for anomaly detection and prognostication used to schedule maintenance. We evaluate the use of model regularizers and constraints to improve the performance of the model and investigate methods to estimate the uncertainty of the models to provide a robust prediction with statistical interoperability. This paper describes the operational experience and known failures of the HVCMs and the proposed ML methodology and the preliminary results of training the AI/ML algorithms.
* G. Dodson, Approach to Reliable Operations, 26-DodsonApproach to Reliable Operation-r1.pdf, Feb., 2010.
NSRRC, Hsinchu, Taiwan