Enhancing integrated navigation system reliability and resilience by decentralized filtering

We present a decentralized filtering architecture for integrated navigation systems with a large number of redundant measurements. The architecture proposed consists of decentralizing the sensor fusion algorithm into a bank of local filters whose output is blended by a master filter using covariance intersection. The problem of isolating a failed sensor is transformed into a problem of excluding failed filters from fusion. Each local filter includes its own fault detection algorithm which is used to alter the weight assigned to the local filters by the master filter. Decentralization contains the impact of a sensor failure in the corresponding local filter and prevents the failure from corrupting the final navigation system output. The performance of the proposed sensor fusion scheme is validated using a simulation study consisting of an INS/Beacon system with 23 sensors. The results from decentralized filters are compared with a centralized filter in the presence and absence of faults. While a centralized filter is seen to have a better accuracy during normal operations, its continuity is compromised when fault occurs. Results show the effectiveness of using simple fault detection and exclusion strategies along with covariance intersection to automatically detect a failed sensor and exclude it in a timely manner. Additional comparative results are presented to highlight the importance of optimal sensor allotment during decentralization. The decentralized filter architecture presented here can support many sensor inputs, hence these algorithms can be used for autonomous systems with a large array of sensors such as multi-constellation GNSS positioning.