Describe a typical fault detection and exclusion (FDE) approach in integrated navigation.

• A. Monitor residuals between predicted measurements and actual sensors; if residuals exceed thresholds, exclude faulty sensor data and reweight.
• B. Always trust GNSS; ignore other sensors.
• C. Randomly discard sensor data.
• D. Only rely on INS.

Answer: (C)

In integrated navigation, a typical fault detection and exclusion approach relies on monitoring the innovations produced by the estimator and using that information to decide when a sensor is faulty. The idea is that each sensor measurement is compared to what the current state estimate predicts, producing a residual or innovation r = z - Hx̂. If a sensor or channel starts behaving badly, its residual will look unlikely given the expected measurement noise, so you test this against a statistical threshold.

Practically, you compute the residual and its expected uncertainty (the residual covariance S = HPHᵀ + R for a linearized case). If the residual is larger than what the threshold allows—often using a chi-square or multi-variance gate—the sensor data are flagged as suspect. Once flagged, you exclude that sensor’s data from the update or downweight its influence (reweighting) so it no longer distorts the state estimate. The filter is then updated with the remaining, trusted measurements, which keeps navigation accuracy and prevents a single bad reading from contaminating the whole fusion.

This approach is applied across sensors in integrated navigation, such as GNSS with an inertial navigation system, where a GNSS measurement that shows an anomalous residual due to multipath or outage can be dropped while the INS continues to propagate, ensuring robustness without abandoning the benefits of fusion. Robust extensions may use adaptive thresholds, multiple hypothesis tests, or robust statistics to handle outliers more gracefully when residuals don’t perfectly follow the assumed noise model.