Fault-Tolerant GPS/INS Navigation System with Application to Unmanned Aerial Vehicle

摘要

A fault-tolerant estimator emerges from the fusion of fault detection and estimation in an architecture centered around a bank of discrete-time observers. Each filter is designed to block the effect on the residual of chosen faults, denoted "nuisances." Using a residual screening method, a fault onset is announced and its source identified. In the bank, one filter is by design insensitive to the identified fault, and therefore, the overall fault-tolerant state estimate reduces to that uncorrupted estimate. The detection performance is independent of the fault signal type (bias, noise, drift, or other), but only rely on the knowledge of how it affects the system model (dynamics and measurement equations). In this sense, this deterministic approach is different from previous voting schemes, which rely on statistical tests. The performance of such a structure is tested on sensor faults. The robustness of the detection / blocking is discussed for turbulence and model mismatch.