Machine learning assisted analytical field calculation of anisotropic cuboid magnets

Analytical expressions for magnetic field calculations offer computing speeds below a millisecond, ideal for magnetic system modelling. However, such analytical models assume perfect homogeneity of the magnetization, which in general is not the case due to the self-demagnetization effect. We introduce a neural network-based algorithm which accurately accounts for the deviations caused by self-demagnetization, while maintaining the extremely low latency of analytical models. To demonstrate the potential of our algorithm for magnetic position and orientation systems, we show that our algorithm is capable of improving the angular accuracy of a joystick with imperfect magnets by an order of magnitude.