Machine Learning on Large Data Streams for CLS Operations

The CLS control system for the particle accelerator and beamlines, which are larger than the size of a football field, generates over 600,000 digital data streams to monitor and control the facility 24/7. The task of monitoring and reacting to these data streams is largely up to the Operators and other Egineers, Physicists and Scientists who manually check the data with the assistance of pre-set thresholds that generate alarms if limits are reached. Often once an alarm is triggered it is too late – the facility has dropped out of operations or equipment has been damaged. This project aims to take the first steps into a new paradigm of predictive maintenance and performance monitoring, by building modern tools to analyse the data for trending and correlations between data streams before an alarm limit is reached. The present method of manual human analysis is very time consuming and the CLS does not have the resources to retrospectively put all their data streams through algorythms that perform the same expert analysis provided by an expert human. This project aims assist the experts and provide a tool to analyse the huge data stream for possible correlations between the data streams that will lead to predicting common failure modes and prevent them from happening.

Faculty Supervisor:

Michael Bradley

Student:

Partner:

Canadian Light Source

Discipline:

Physics

Sector:

Education; Technology; Artificial Intelligence

University:

University of Saskatchewan

Program: