The increasing autonomy of spacecraft demands fault-detection systems that are both reliable and explainable. This work addresses eXplainable Artificial Intelligence for onboard Fault Detection, Isolation and Recovery within the Attitude and Orbit Control Subsystem by introducing a framework that enhances interpretability in neural anomaly detectors. We propose a method to derive low-dimensional, semantically annotated encodings from intermediate neural activations, called peepholes. Applied to a convolutional autoencoder, the framework produces interpretable indicators that enable the identification and localization of anomalies in reaction-wheel telemetry. Peepholes analysis further reveals bias detection and supports fault localization. The proposed framework enables the semantic characterization of detected anomalies while requiring only a marginal increase in computational resources, thus supporting its feasibility for on-board deployment.
Figures & tables
Figure 1: Block scheme of the proposed framework extracting peephole vectors p from a target layer involving three stages: dimensionality reduction (DR), statistical characterization (SC), and semantic mapping (SM).
GWN
Offset
Impulse
PSA
Step
XI′
1
1
1
1
1
XII′
1
0.97
1
1
1
Table 1: AUC values for the five types of synthetic anomalies.
Figure 2: Block diagram of the autoencoder producing anomaly score, including semantic extraction from the encoder’s last layer
Figure 3: Anomaly identification when applied to all channels or to a single RW .
Figure 4: Confusion matrices for RW identification using Peephole for five anomalies.
Figure 5: Telemetries of a true anomalous event along with the autoencoder’s score and the corresponding peepholes vectors visualized as a heatmap.