A NARXNN-based Fault-Tolerant Method for IMU-based Multi-source Integrated Navigation System

• Published in: Chinese Control Conference pp. 4961 - 4968
• Main Authors: Yan, Yuwei, Yang, Jing
• Format: Conference Proceeding
• Language: English
• Published: Technical Committee on Control Theory, Chinese Association of Automation 28.07.2024
• Subjects: Accuracy; Fault tolerance; Fault tolerant systems; Global navigation satellite system; Integrated navigation system; Magnetic sensors; Multi-source; NARX neural network; Navigation; Systems operation
• ISSN: 1934-1768
• DOI: 10.23919/CCC63176.2024.10662805

Integrated navigation systems typically consist of an Inertial Measurement Unit (IMU) and several auxiliary navigation sensors, which are the primary scheme for achieving high-precision and strong-reliability positioning information. External factors like tunnel occlusion or electromagnetic interference, as well as internal factors such as device aging, can result in reduced signal quality or even failure of auxiliary navigation sensors during system operation. To enhance the precision of the integrated navigation system under these factors, this paper proposes a fault-tolerant method based on Nonlinear AutoRegressive with eXogenous input Neural Network (NARXNN) to construct a measurement reference system for substituting for the fault sensors in federal Kalman filter. By employing an integrated navigation system consisting of Global Navigation Satellite System (GNSS), odometry and magnetometer as the subject of analysis, simulation experiments are conducted to verify the performance of the proposed method. The results demonstrate that the method can effectively predict the sensor measurements in short-time sensor fault period. When the fault is a ramp fault, or the fault is a step fault while the information provided by the faulty sensor is non-redundant, the proposed method can effectively enhance the accuracy of the integrated navigation system when the measurement signal quality decreases due to sensor failure. Compared to isolating the fault sensor, the proposed method results in smaller mean modulus of the estimation errors.