Development of a Physics-Informed Neural Network and Magnetic Field Sensor System for Real-Time Determination of Plasma Magnetic Field

Fusion research is rapidly advancing toward high-performance magnetic-confinement devices, including fusion pilot and power plants, in which plasma sustainment depends on the structure of the confining magnetic field. Two methods commonly used to determine the local magnitude and topology of the magnetic field are plasma-edge probes and electron beam imaging, both of which require physical probes in the vacuum vessel. Access to and within advanced fusion devices will be severely limited and these established methods of magnetic field determination will be untenable. We propose to develop a novel magnetic field diagnostic system, B-Swarm, that combines a physicsinformed neural network with a set of three-axis magnetic field sensors. While B-field probes are not new, our approach of using them with a physics-informed neural network is novel. Quarter-dollar sized B-field sensors (Bees) will be placed on the surface of the vacuum vessel. A computer (the Hive) will use serial communication protocols to acquire local B-field measurements from the sensors. The physicsinformed neural network running on the computer will predict the most likely magnetic field, Bvessel, that satisfies both Maxwell’s equations and the B-field measurements. This prediction may be accomplished using a combination of open-source machine learning libraries, and new custom algorithms. Advantages of our innovative B-swarm, relative to insertable probes and imaging systems, are its location (external vs internal to vacuum vessel), serviceability (due to ease of access), real-time determination of magnetic field, and deployability (where traditional probes are not feasible). Target customers for B-Swarm will vary with time with (a) immediate customers being facilities designing and verifying magnet coil optimization for near-term devices, and (b) fusion pilot plants that will use the system for operational stability. Key features that make it attractive to a wide range of customers, from university research laboratories to private fusion companies, are its customizability, which enables deployment on devices of varying size, and its modularity which makes it readily maintainable and upgradeable. Markets existing beyond fusion science are fields of particle accelerators, medical imaging, avionics and space. During Phase-1, Xantho Technologies will: (1) develop a prototype B-Swarm system (having up to 100 sensors), and (2) evaluate its ability to accurately determine Bvessel on existing, readily accessible magnetic confinement devices such as the Helically Symmetric eXperiment. During Phase-2, we will: (a) advance the system’s robustness to radiation exposure, (b) tackle challenges of performance for devices with a range of low to substantial plasma current, and (c) develop algorithms that utilizes known limits for acceptable magnetic field components and topology. When deployed, B-Swarm will aid in real-time operation of high-power fusion devices and prediction of plasma disruptions (key to their safeguarding).