A Giant-Step Baby-Step Classifier For Scalable and Real-Time Anomaly Detection In Industrial Control Systems and Water Treatment Systems

Real-time anomaly detection is achievable with millisecond response times: The paper demonstrates that their method can detect anomalies in industrial control systems in milliseconds, which is critical for preventing safety incidents in facilities like water treatment plants where rapid response is essential.

Explainability and speed are now compatible: Unlike previous AI/ML approaches, this method achieves both fast anomaly detection (millisecond response) AND explainability simultaneously—meaning operators can understand not just that something is wrong, but exactly which sensors or controls are causing the problem.

The method pinpoints specific failure sources: The algorithm can identify the exact sensor(s) and actuator(s) responsible for detected anomalies, enabling faster troubleshooting and more targeted responses rather than system-wide shutdowns.

High accuracy with practical applicability: The system achieved 97.72% accuracy in detecting deviations, suggesting it could be reliably deployed in real industrial settings like water treatment facilities to catch attacks, equipment faults, and unexpected system states.

Simplified approach may be more trustworthy: By using linearization of sensor-actuator relationships rather than complex "black box" AI models, the method creates a more transparent and auditable system—important for critical infrastructure where understanding how safety decisions are made is paramount.