PKBPNN-Based Tracking Range Extending Approach for TMR Magnetic Tracking System

Abstract

The magnetic tracking system, which is based on a permanent magnet and a magnetometer array, has numerous potential applications in the biomedical and industrial area. However, its tracking accuracy drops off sharply with the increase of tracking distance because of both the interference of environmental magnetic field and sensor measurement noise. To extend the magnetic tracking range, two novel methods were proposed in this study. First, the state-of-the-art tri-axial tunnel magnetoresistance (TMR) sensors, which possess the most significant field sensitivity and signal-to-noise ratio (SNR) over other types of magnetoresistive sensors, were adopted to construct the sensor array. Second, a fusion approach was proposed for tracking range extending. The particle swarm optimization-Levenberg Marquardt (PSO-LM) method based on the magnetic dipole model was applied in the near field (i.e., near-source zone), and the prior knowledge based back propagation neural network (PKBPNN) was adopted to the far field (i.e., far-source zone). In the transition zone, these two algorithms were fused by using an adaptive sigmoid function. The trained artificial neural network (ANN) model embodied the physical model errors, the sensor installation errors, and the inherent characteristics of adopted magnetometers. Therefore, it has greater tracking performance than singly using the PSO-LM in the far-source zone. The experimental results show that the tracking errors decrease from (18.24 ± 9.37mm, 12.45 ± 3.37°) to (8.95 ± 1.74mm, 7.97 ± 2.08°) in the tracking range between 216 and 296 mm. Besides, the tracking distance is extended to 396 mm, with the position error of less than 25 mm. It can be concluded that this approach has significantly extended the tracking range of the magnetic tracking system.