Explainable AI guided unsupervised fault diagnostics for high-voltage circuit breakers

Real-world fault detection without labeled data: The paper addresses a critical gap in AI applications for infrastructure—most existing systems require labeled examples of known faults to work, but real-world circuit breakers fail in unpredictable ways. This framework can detect faults using only examples of healthy equipment, making it more practical for actual deployment.

Continuous monitoring without shutting down equipment: Current monitoring systems require circuit breakers to be disconnected from the power grid to check physical parameters like gas pressure. This new approach uses vibration and acoustic signals to monitor equipment while it remains operational, reducing downtime and improving safety.

Making AI decisions interpretable for domain experts: The framework incorporates explainable AI (XAI) techniques, meaning engineers and technicians can understand why the system flagged a particular fault rather than just receiving a black-box warning. This transparency is essential for building trust in AI systems managing critical infrastructure.

Combining detection with diagnosis in one system: Rather than separate steps, the framework both detects that something is wrong and attempts to identify what type of fault occurred, streamlining the diagnostic process for maintenance teams.

Potential to prevent catastrophic failures: By enabling early detection of circuit breaker degradation through non-intrusive monitoring, this approach could help prevent unexpected equipment failures that could disrupt power grids or create safety hazards.