PhaseNet++: Phase-Aware Frequency-Domain Anomaly Detection for Industrial Control Systems via Phase Coherence Graphs
PhaseNet++ introduces a novel autoencoder designed to identify anomalies within industrial control systems (ICS) by emphasizing frequency domain analysis. Utilizing the Short-Time Fourier Transform (STFT) with sliding sensor windows, it preserves essential phase and magnitude information. A distinctive component, the Phase Coherence Index (PCI), draws from neuroscience to measure phase consistency across frequency bins, creating a continuous adjacency matrix. This matrix is processed through a graph attention network, effectively overcoming the shortcomings of existing methods that overlook phase data in time-frequency analysis. The research is available on arXiv under ID 2605.00929.
Key facts
- PhaseNet++ is a frequency-domain autoencoder for ICS anomaly detection.
- It uses STFT of sliding sensor windows, retaining magnitude and phase spectra.
- Phase Coherence Index (PCI) is inspired by Phase Locking Value from neuroscience.
- PCI summarizes pairwise phase consistency into a continuous adjacency matrix.
- The matrix guides a graph attention network.
- State-of-the-art methods discard phase information from time-frequency transformations.
- Phase information is argued to be a complementary detection modality.
- Paper ID: arXiv:2605.00929.
Entities
Institutions
- arXiv