Using machine learning to maximise impact of fusion energy experimental facility virtual twin
Lead Research Organisation:
Swansea University
Department Name: College of Engineering
Abstract
The 'Heat by Induction to Verify Extremes' (HIVE) facility at UKAEA is used to test components under the extreme environments experienced in a fusion device. To maximise the value of experimental data the candidate will work as part of a team (involving partners from academia, our national institutes (e.g. STFC) and industrial partners) to build Digital Twin capability around the UKAEA HIVE Facility. Each time an experiment is performed the developed platform will use the input parameters to auto-trigger an equivalent simulation running the twin through the same duty cycle as the real apparatus. A scheme will be developed to auto-compare experiment and simulation to warn if equipment needs recalibrating or if large discrepancies are present, and there is confidence in the experiment, it can alert that the model is not accurately capturing all the mechanisms of the testing regimes.
A machine learning framework will then be developed to query the database of experiment/simulation results built up over time. The aim would be to use this for steering which future tests to perform (reducing the number required), optimising experimental run time (again, by identifying the optimal parameter settings), anomaly detection and predictive maintenance. Ultimately the goal is to turn a much larger fraction of the data recorded into invaluable information. We are now in an era where all previous experiments should form a "prior" for future experiments and operational planning - no longer does it make sense to employ "experts" and to rely upon their memories, experience and gut feeling in the way we do our experiments and model verification. Research Areas - Continuum mechanics, materials for energy applications, artificial intelligence technologies, Engineering design, UK magnetic fusion research programme.
A machine learning framework will then be developed to query the database of experiment/simulation results built up over time. The aim would be to use this for steering which future tests to perform (reducing the number required), optimising experimental run time (again, by identifying the optimal parameter settings), anomaly detection and predictive maintenance. Ultimately the goal is to turn a much larger fraction of the data recorded into invaluable information. We are now in an era where all previous experiments should form a "prior" for future experiments and operational planning - no longer does it make sense to employ "experts" and to rely upon their memories, experience and gut feeling in the way we do our experiments and model verification. Research Areas - Continuum mechanics, materials for energy applications, artificial intelligence technologies, Engineering design, UK magnetic fusion research programme.
