Graph learning improves detection of anomalies in industrial multivariate data
GSLAD: Prototype-Regularized Graph Structure Learning for Multivariate Time Series Anomaly Detection
Machine Learning
Summary
Detecting anomalies in complex systems can be tricky when individual measurements look normal but relationships between them change. The authors present a method called GSLAD that learns patterns of these relationships, grouping normal operation types and spotting when the system’s structure shifts unusually. By comparing current observations to these typical patterns, the method better finds and explains faults in industrial data. Experiments show GSLAD is effective across multiple real-world datasets.
What this means in practice
- •For industrial monitoring teams: Improve fault detection by using learned structural patterns of sensor relationships to find subtle anomalies in machine operations.
- •For data engineers in manufacturing: Implement condition-aware graph models to enhance predictive maintenance systems by capturing regime-specific operational modes.
Authors
Zepeng Zhang, Fuad Khuri, Keivan Faghih Niresi, Olga Fink
Abstract
Unsupervised multivariate time series anomaly detection methods typically identify anomalies through forecasting, reconstruction, or representation discrepancies. However, industrial faults may first alter inter-variable structural patterns while individual trajectories remain close to normal, resulting in weak anomaly signals. In this paper, we propose GSLAD, a prototype-regularized graph structure learning framework that uses structural deviations for anomaly scoring. GSLAD adopts a two-phase training strategy. First, a condition-aware graph learner and a graph-based forecaster are optimized with predictive supervision. The inferred normal graphs are then clustered into multiple structural prototypes representing different normal operating regimes, with edge-wise variability characterizing structural uncertainty. Deviations from these prototypes regularize the graph learner in the second phase, encouraging stable and regime-specific structural patterns. During inference, uncertainty-normalized structural deviation is combined with predictive deviation for anomaly scoring. Experiments on four industrial benchmarks demonstrate strong overall performance of GSLAD and confirm the effectiveness of structural deviation for anomaly detection and diagnosis.