Abstract
State estimation and fault diagnosis are central to the safe operation of unmanned aerial vehicles (UAVs) whose sensors and actuators may degrade in flight. Conventionally the estimator holds its covariances fixed while the fault detection and isolation (FDI) logic sits outside it, so neither informs the other and robustness degrades when both fault types occur. A robust adaptive active inference (RAAIF) framework closes this loop. From a discrete-time linear–Gaussian generative model the nominal estimator is derived as a precision-weighted prediction–correction update, then extended by two bounded covariance-matching mechanisms: the observation side adapts from innovation statistics under sensor degradation, the process side from correction-residual statistics under actuator-induced mismatch. Both responses, with the channel-resolved residual energies, are read by a hidden Markov model over healthy, sensor-fault, actuator-fault and concurrent-fault modes, whose posterior gates the covariances of the next cycle. The exact joint law of the two residuals is derived, the estimation error is shown to be mean-square bounded under the projected adaptation, and a detectability index is obtained for the signatures covariance matching separates. Sequential and concurrent mixed-fault simulations on a reduced UAV rotational benchmark compare RAAIF with nominal discrete active inference (NDAIF), a robust adaptive optimal Kalman filter (RAOKF) and three further adaptive and multiple-model baselines. RAAIF resolves each fault to its channel alongside the state estimate, and the posterior gate raises concurrent attribution by half again. The fault-window error falls against NDAIF in both protocols, and the measured thresholds and false-alarm rates match the predicted values.
| Original language | English |
|---|---|
| Article number | 114079 |
| Journal | Aerospace Science and Technology |
| Volume | 180 |
| DOIs | |
| State | Published - Jan 2027 |
Keywords
- Active inference
- Covariance matching
- Fault detection and isolation
- Robust adaptive estimation
- UAV rotational dynamics
Fingerprint
Dive into the research topics of 'Robust adaptive active inference for resilient UAV state estimation under sensor and actuator faults'. Together they form a unique fingerprint.Cite this
- APA
- Author
- BIBTEX
- Harvard
- Standard
- RIS
- Vancouver