Learning faults in time: sequential behavioural modelling for complex fault detection in multi-robot systems

Reliable fault detection in multi-robot systems requires models capable of capturing complex, time-dependent fault signatures that manifest over extended temporal horizons rather than instantaneous observations alone. Existing data-driven approaches operate reactively on behavioural snapshots, failing to capture fault modes whose discriminative signature depends on temporally ordered precursors. This work formalises a theoretical impossibility result demonstrating that memoryless classifiers are provably insufficient, and presents a Long Short Term Memory (LSTM) model operating on generic, primitive behavioural features to learn these latent temporal dependencies without domain-specific priors. Two complex fault types are evaluated: progressive actuator degradation and intermittent actuator dropout, each paired with a distinct swarm movement pattern representing varying levels of temporal complexity. Experimental results demonstrate that sequential models are essential when the fault signature is defined by ordered precursors, while memoryless models partially suffice for simpler temporal structures. Compute benchmarks confirm real-time feasibility for distributed onboard deployment.