Stony Brook University, Stony Brook, New York, USA
BNL, Upton, Long Island, New York, USA
Bayesian Analysis provides a statistical framework for updating prior knowledge as observational evidence is acquired. It can handle complex and realistic models with flexibility. The Beam Permit System (BPS) of RHIC plays a key role in safeguarding against the faults occurring in the collider, hence directly impacts RHIC availability. Earlier a multistate reliability model* was developed to study the failure characteristics of the BPS that incorporated manufacturer and military handbook data. Over the course of its 15 years of operation, RHIC has brought forth operational failure data. This work aims towards the integration of earlier reliability calculations with operational failure data using Bayesian analysis. This paper discusses the Bayesian inference of the BPS reliability using a two-parameter Weibull survival model, with unknown scale and shape parameters. As the joint posterior distribution for Weibull with both parameters unknown is analytically intractable, the Markov Chain Monte Carlo methodology with Metropolis-Hastings algorithm is used to obtain the inference. Selection criteria for the Weibull distribution, prior density and hyperparameters are also discussed.
*P. Chitnis et al., 'A Monte Carlo Simulation Approach to the Reliability Modeling of the Beam Permit System of Relativistic Heavy Ion Collider (RHIC) at BNL', Proc. of ICALEPCS'13, San Francisco, CA.
Stony Brook University, Stony Brook, New York, USA
BNL, Upton, Long Island, New York, USA
The Quench Detection System (QDS) of RHIC detects the Superconducting (SC) magnet quenches by voltage sensing. The real-time voltage across the SC magnet is compared with a predicted voltage from a behavioral model, a deviation from which triggers the quench event and energy extraction. Due to the limitations of the magnet model, many false quench events are generated that affect the RHIC availability. This work is targeted towards remodeling the magnets through nonlinear system identification for the improvement in QDS reliability. The nonlinear electrical behavior of the SC magnets is investigated by statistical data analysis of magnet current and voltage signals. Many data cleaning techniques are employed to reduce the noise in the observed data. Piecewise regression has been used to examine the saturation effects in magnet inductance. The goodness-of-fit of the model is assessed by field testing and comprehensive residual analysis. Finally a new model is suggested for the magnets to be implemented for more accurate results.
Stony Brook University, Stony Brook, New York, USA
BNL, Upton, Long Island, New York, USA
The RHIC Beam Permit System (BPS) monitors the anomalies occurring in the collider and restores the machine to a safe state upon fault detection. The reliability of the BPS thus directly impacts RHIC availability. An analytical multistate reliability model of the BPS has been developed to understand the failure development and propagation over store length variation. BPS has a modular structure. The individual modules have joint survival distributions defined by competing risks with exponential lifetimes. Modules differ in functionality and input response. The overall complex behavior of the system is analyzed by first principles for different failure/success states of the system. The model structure changes according to the type of scenario. The analytical model yields the marginal survival distribution for each scenario versus different store lengths. Analysis of structural importance and interdependencies of modules is also examined. A former Monte Carlo model* is used for the verification of the analytical model for a certain store length. This work is next step towards building knowledge base for eRHIC design by understanding finer failure characteristics of the BPS.
*P. Chitnis et al., 'A Monte Carlo Simulation Approach to the Reliability Modeling of the Beam Permit System of Relativistic Heavy Ion Collider (RHIC) at BNL', Proc. ICALEPCS'13, San Francisco, CA.
BNL, Upton, Long Island, New York, USA
This past year the Controls group within the Collider Accelerator Department at Brookhaven National Laboratory replaced the Network Attached Storage (NAS) system that is used to store software and data critical to the operation of the accelerators. The NAS also serves as the initial repository for all logged data. This purchase was used as an opportunity to categorize the data we store, and review and evaluate our storage strategies. This was done in the context of an existing policy that places no explicit limits on the amount of data that users can log, no limits on the amount of time that the data is retained at its original resolution, and that requires all logged data be available in real-time. This paper will describe how the data was categorized, and the various storage strategies used for each category.
BNL, Upton, Long Island, New York, USA
Analysis of accelerator data has traditionally been done using custom tools, either developed locally or at other laboratories. The actual data repositories are openly available to all users, but it can take significant effort to mine the desired data, especially as the volume of these repositories increases to hundreds of terabytes or more. Much of the data analysis is done in real time when the data is being logged. However, sometimes users wish to apply improved algorithms, look for data correlations, or perform more sophisticated analysis. There is a wide spectrum of desired analytics for this small percentage of the problem domains. In order to address this tools have been built that allow users to efficiently pull data out of the repositories but it is then left up to them to post process that data. In recent years, the use of tools to bridge standard analysis systems, such as Matlab, R, or SciPy, to the controls data repositories, has been investigated. In this paper, the tools used to extract data from the repositories, tools used to bridge the repositories to standard analysis systems, and directions being considered for the future, will be discussed.
BNL, Upton, Long Island, New York, USA
Applications used for Collider-Accelerator Controls at Brookhaven National Laboratory typically run as console level programs on a Linux operating system. One essential requirement for accelerator controls applications is bidirectional synchronized IO data communication. Several new web frameworks (Vaadin, GXT, node.js, etc.) have made it possible to develop web based Accelerator Controls applications that provide all the features of console based UI applications that includes bidirectional IO. Web based applications give users flexibility by providing an architecture independent domain for running applications. Security is established by restricting access to users within the local network while not limiting this access strictly to Linux consoles. Additionally, the web framework provides the opportunity to develop mobile device applications that makes it convenient for users to access information while away from the office. This paper explores the feasibility of using the Vaadin web framework for developing UI applications for Collider-Accelerator controls at Brookhaven National Laboratory.