Fault Detection and Isolation

Fault Detection and Isolation (FDI) is the process of detecting when a system component has failed, determining which component failed, and isolating it so the system can adapt.

In AEGIS

The FDI module serves as the system’s “diagnostic nervous system.” It continuously monitors the health of the engine cluster and triggers contingency responses when failures are detected.

Detection Logic

1. Expected acceleration: Computed from commanded throttle and known vessel mass
2. Measured acceleration: Provided by the State Estimator (EKF)
3. Comparison: If deviation exceeds FDI_THRESHOLD = 3.0 for 50 consecutive ticks, a fault is declared

Isolation

Once a fault is detected, the FDI brute-forces failure combinations to identify which specific engine has failed. It tests all possible subsets:

• Single engine failure
• Double engine failure
• etc.

It matches the predicted acceleration signature of each combination against the measured deviation to isolate the failed engine(s).

Contingency Response

Fault Severity	Response
Single engine failure	FDI flags engine, allocator remaps around it
2+ simultaneous failures	HARD_ABORT
Degenerate allocation	HARD_ABORT
DT spike (game lag)	Skip KF predict, hold FDI, guidance still runs

Future: NN-ADRC Integration

The planned NN-ADRC will change FDI logic:

• Monitor ESO disturbance estimate z₃ for sudden spikes (discrete actuator failure)
• Monitor NN compensatory output Δr̈ for persistent non-zero values (permanent engine loss)

Related Concepts

• Control Allocation — Remaps around failed engines
• State Estimation — Provides measured acceleration
• Extended State Observer — Future FDI monitoring source
• Neural Network Compensator — Masks acceleration deviations
• Redundancy Management — System-level fault tolerance

Sources

• AEGIS Project (src/fdi/fdi.py)