Automated Machine Learning Workflows for Fusion Power Plant Design

The need to meet increasing global energy demand and also address 2050 net zero targets is placing fusion energy in the spotlight. The authors are investigating the ad- vances in digital technology necessary to deliver a fusion power plant by 2040. We are currently evaluating the use of the Nvidia Omniverse platform for engineering, with the specific need to consider fusion power plants as a whole system. In a plant that uses a magnetically confined plasma, fusion generates neutrons which pass through the whole machine. The design process needs to consider how to protect some sys- tems from the neutrons, whilst in other parts of the machine, the neutrons can be used to generate new fuel. It is difficult to compartmentalise the design process for these two opposing requirements. Therefore a requirement for conceptual design is the ability to carry out fast physics-informed simulations for many coupled systems at the same time. This paper describes how the authors have integrated the Galaxy workflow engine with the Omniverse, automating the execution of a suite of containerised open source software applications that can be used for training surrogate models. Automa- tion is essential if AI and machine learning is to be leveraged in the design of complex engineering systems.