Neural Network-Assisted Magnetic Moment Measurement Using an Atomic Magnetometer

作者：Zhongyu Wang, Jixi Lu, Ziao Liu, Xiaoyu Li, Jianwei Sheng, Jianli Li · 发表于：IEEE Transactions on Instrumentation and Measurement · 年份：2025 · DOI：10.1109/tim.2025.3544322 · 被引用次数：5 · 研究领域：Magnetic Field Sensors Techniques、Atomic and Subatomic Physics Research、Inertial Sensor and Navigation

The accurate measurement of magnetization is crucial in paleomagnetism, paleontology, and materials science. This study proposes a novel approach for quantifying magnetization based on an atomic magnetometer. First, a hybrid model that combines the analytical and magnetic dipole models is established to represent the magnetic source. The spatially varying magnetic field induced by sample scanning is measured at specific positions using a spin-exchange relaxation-free (SERF) atomic magnetometer. This enables the extraction of the magnetic moment information from the source through a nonlinear fitting for an accurate calculation of magnetization. We incorporate the Bayesian optimization long- and short-term time-series network (BO-LSTNet) model to efficiently learn from a limited number of measurement outcomes. By jointly optimizing the accuracy and dynamic range of magnetic moment measurement, the model can quickly determine the optimal measurement position for magnetic sources of different sizes. This method achieves the highest measurement accuracy while ensuring the dynamic range of magnetic moment measurement. Our results demonstrate a static sensitivity of$1.6\times 10^{-6}$emu, a dynamic sensitivity of$1.9\times 10^{-7}$emu/Hz1/2 at 20–80 Hz, and a dynamic range of up to$6.6\times 10^{-4}$emu. By employing the proposed method, limited measurement data are effectively utilized to automatically determine optimal measurement positions for the coils of various sizes. Experim...