Navigation Sensor Failure Detection without Sensor Redundancy

For the reliable control of unmanned aerial and underwater vehicles, a robust attitude solution is needed in order to control the vehicle correctly along its three axes. Emerging sensor faults endanger the attitude estimates, and thus the control of the vehicle, and must be detected as early as possible in order to counteract and mitigate, or possibly even prevent, any negative effects. A common approach to detect erroneous sensor data is the use of sensor redundancies in which measurements of three or more sensors are compared. However, this concept is not applicable in all cases, especially with regard to unmanned vehicles that are often limited by size, weight, cost and other constraints. This paper describes an algorithm, developed to detect faults and degradation in navigation sensor data for an Autonomous Underwater Vehicle (AUV) with a limited navigation sensor suit, i.e. without sensor redundancies. The algorithm aims to calculate an integrity value for each sensor channel, based on the sensor faults or degradations detected. On the one hand, the calculated integrity data is meant to be used by Bayes filter based navigation algorithms (e.g. Kalman Filter) to adjust their confidence in the measured data used to correct the state predictions. On the other hand, the integrity values are used to monitor the overall health condition of the navigation sensor suit, enabling downstream software modules to come to decisions regarding the continuation of the mission.