Innovation-Sequence-Based Adaptive Kalman Filtering and Fault Detection for Integrated Navigation

To address the challenge of detecting small-amplitude, slowly drifting faults that are difficult to separate from nominal measurements in integrated navigation, this paper proposes an innovation-sequence–driven adaptive Kalman filter (ISAKF). The method evaluates the mean and variability of the innovation over short-to-medium time horizons to construct adaptive measurement weights, thereby softly down-weighting suspicious channels. It preserves the standard Kalman filtering structure, requires no fault priors, and can cooperate with conventional detectors such as Normalized Innovation Squared (NIS), CUSUM, and SPRT. INS/GNSS integrated-navigation simulations show that, under nominal conditions, ISAKF achieves accuracy comparable to the conventional Kalman filter (KF) and the Sage–Husa adaptive Kalman filter (SHAKF); when an altitude-channel step bias occurs, ISAKF promptly detects and isolates the fault, reducing a measurement offset of approximately 20 m to within 1 m; under a 0.3 m/s slowly varying fault, it disconnects the faulty measurement in about 2 s, limits the maximum estimation bias to roughly 0.6 m, and drives the associated statistics reliably beyond their thresholds; under strong-noise faults, ISAKF performs on par with SHAKF. These results indicate that ISAKF, with low implementation complexity, markedly improves the detectability and isolation timeliness for weak, slowly varying faults while maintaining robustness and nominal-condition accuracy.