Diagnosis driven Anomaly Detection for Cyber-Physical Systems

In Cyber-Physical Systems (CPS) research, the detection of anomalies-identifying abnormal behaviors-and the diagnosis-pinpointing the underlying root causes-are frequently considered separate, isolated tasks. However, diagnostic algorithms necessitate symptoms, i.e., temporally and spatially isolated anomalies, as inputs. Therefore, integrating anomaly detection and diagnosis is essential for developing a comprehensive diagnostic solution for CPS. This paper introduces a method leveraging deep learning for anomaly detection to effectively identify and localize symptoms within CPS. Our approach is validated on both simulated and real-world CPS datasets, demonstrating robust performance in symptom detection and localization when compared to other state-of-the-art models. Copyright (c) 2024 The Authors.