A machine learning based approach to active magnetic field cancellation

The Neutron Electric Dipole Moment experiment at the Spallation Neutron Source

(nEDM@SNS) is being commissioned to measure the nEDM to a precision of 1.6 ×

10^-28 e·cm in the critical spin dressing mode. Achieving this level of precision

requires extreme control over magnetic fields in the measurement region, with

strict requirements on field gradients. The experiment features many layers of

passive magnetic shielding, including a superconducting lead shield. It is

important for there to be zero field inside the lead shield when it goes

superconducting, otherwise magnetic flux will be trapped. To aid in this, an active

magnetic field cancellation system will be used to monitor and cancel ambient

fields. A prototype system is being developed for the Systematics and Operational

Studies apparatus at the PULSTAR reactor (SOS@PULSTAR) at North Carolina

State University. SOS@PULSTAR provides a testbed for the nEDM@SNS

experiment, sharing many of the same design features but its smaller size allows

for more rapid warming/cooling cycles. The strict magnetic field uniformity

requirements for nEDM@SNS and SOS@PULSTAR prevent a magnetic field probe

from being placed inside the superconducting shield where the field needs to be

zeroed. To overcome this, a machine learning based method is being developed

where external sensors predict the field in the region of interest. This predicted

value can then be used as feedback for the active cancellation system.