Predictive Model-Based Correction of Magnetic Sensor Array Sway Errors

Abstract

Magnetic sensor arrays are typically used to detect magnetic targets. Currently, research on sensor array calibration focuses on solving the problems of inconsistent sensitivity, zero offset, and non-orthogonality in individual sensors, and misalignment errors between sensors. However, in magnetic field detection, sensor arrays are usually mounted on a platform or carried as a handheld device and are prone to random swaying during the detection process, leading to changes in the attitude and position of the magnetic sensors, which in turn generates swaying errors. In this study, the source of swaying error in sensor arrays was first analyzed theoretically, and a swaying error model of a magnetic sensor was established. Second, a swaying error calibration method was proposed in combination with the classical prediction model—Gaussian process regression (GPR), backpropagation (BP) neural network, and support vector machine (SVM) in the field of artificial intelligence. The experimental results show that the prediction performance based on the BP neural network is the most outstanding. After correction, the relative error percentage of the swaying error in the magnetic field data decreased significantly from 165.50% to 9.35%, which is a significant improvement of the correction effect. In addition, we conducted model comparison experiments in different environments, and the results show that the BP model performs well in various environments, demonstrating its strong generalization ability and robustness. Finally, the distance error of the magnetic dipole position was significantly reduced after calibration, from 1.56, 1.12, and 2.50 m to 0.17, 0.04, and 0.04 m, respectively. Thus, the effectiveness of the calibration method was verified.