Real-time sensor fault detection and compensation in a passive magnetic field-based localization system

In modern mechatronic systems, due to the large number of electronic sensors employed, it is inevitable to encounter sensor faults over time. Taking the localization system for medical intervention as an illustrative example, high localization accuracy is usually the highest priority but the system should also be resilient to occasional sensor faults and failures. Presence of sensor faults, especially those irreversible ones, could greatly deteriorate the localization performance and in turn jeopardize patient safety. In this paper, inspired by current research and development of a novel real-time localization and monitoring system for nasogastric intubation, the effects of sensor faults that can occur in such magnetic field-based localization systems were first explored and classified. Three types of common sensor faults were identified, and the corresponding real-time detection methods were discussed. It focused on the two most commonly-used localization approaches, the inverse optimization method and the direct ANN method, and proposed real-time compensation remedies for both the methods in presence of such sensor faults. The feasibility and efficacy of the proposed compensation remedies were examined by numerical simulations as well as real experimental data. It is shown in presence of a single sensor fault with excessive noise, the localization performance can be deteriorated by up to 300%. But through implementation of the proposed compensation scheme, which does not sacrifice computational time (less than 1%), the inherent redundancy of the system could be used to achieve similar localization performance as that under normal conditions. The work presented in this paper aims to improve the robustness and reliability of multisensor-based localization or tracking systems.