Remote-sensing based control of 3D magnetic fields using machine learning for in-operando applications

TL;DR

This paper presents a machine learning approach to precisely control 3D magnetic fields in in-operando experiments, overcoming indirect measurement challenges and enabling complex magnetic state studies at micro- and nano-scales.

Contribution

It introduces a neural network-based calibration method for magnetic field control using a hexapole electromagnet, improving accuracy and generalization in indirect measurement scenarios.

Findings

Neural network achieves high calibration accuracy.

Method effectively generalizes to unseen magnetic sequences.

Enables precise 3D magnetic control in constrained setups.

Abstract

In-operando techniques enable real-time measurement of intricate physical properties at the micro- and nano-scale under external stimuli, allowing the study of a wide range of materials and functionalities. In nanomagnetism, in-operando techniques greatly benefit from precise three-dimensional (3D) magnetic field control, enabling access to complex magnetic states forming in systems where multiple energies are set to compete with each other. However, achieving such precision is challenging and uncommon, as specific applications impose constraints on the type and geometry of magnetic field sources, limiting their capabilities. Here, we introduce an approach that leverages machine learning algorithms to achieve precise 3D magnetic field control using a hexapole electromagnet that is composed of three independent, non-collinear dipole electromagnets. In our experimental setup, magnetic field sensors are placed at a distance from the sample position due to inherent constraints, leading to indirect field measurements that differ from the magnetic field experienced by the sample. We find that the existing relationship between the remote and sample frames of reference is non-linear, thus requiring a more complex calibration method. To address this, we employ a multi-layer perceptron neural network that processes multiple inputs from a dynamic magnetic field sequence, effectively capturing the time-dependent non-linear field response. The network achieves high calibration accuracy and demonstrates exceptional generalization to unseen magnetic field sequences. This study highlights the significant potential of machine learning in achieving high-precision control and calibration, crucial for in-operando experiments where direct measurement at the point of interest is not possible.