A Method for Multiple Fault Detection and Isolation of Redundant Inertial Navigation Sensor Configurations

Abstract

A method for fault detection and isolation of sensors in an inertial navigation system is presented, with capability of detecting several simultaneous faults of different magnitude. The method is based on projecting the measurement vector of sensor signals onto the orthogonal complement of the range space of the system matrix. The system matrix is given by the geometrical properties of sensor placements. A theorem is presented, proving that the method will detect and isolate several simultaneous faults, given that the axes of any n sensors of the full sensor set are linearly independent, where n is the navigation dimension. The theorem is given under the assumption of exact arithmetic and noise-free measurement. Practical algorithms are given based on the theorem and the QR-factorisation of the system matrix. The algorithms are tested on simulated inertial navigation system data. I. INTRODUCTION In an inertial measurement unit (IMU) a number of sensors are used to measure the two vector quantities acceleration and angular rate, of the IMU relative to an inertial reference frame. In a redundant inertial measurement unit (RIMU) the measure- ments are done by more sensors than are actually necessary, see Fig. 1. The use of extra sensors introduces additional complexity, but offers improved accuracy and robustness. Accuracy because the estimation of a navigation solution can be based on more, independent, measurements. Robustness because additional measurements makes fault detection and sensor isolation possible. RIMU:s are therefore used in in- ertial navigation systems and other applications where high availability, robustness and accuracy is required. In this report a method for fault detection and sensor isolation (FDI) is studied, which can isolate several sensor faults simultaneously.