UMAD-G: Unsupervised Multi-Modal Time Series Anomaly Detection via Graph

In the wake of Industry 4.0, many companies are equipping their systems with more sensors and collecting huge amounts of sensor data. This data, in the form of time series, plays an important role in increasing productivity and implementing smart factories. One of the most important things is to detect anomalies caused by various factors (equipment ageing, sudden stops in the production line, interference from external factors). One of the limitations of existing time series anomaly detection research is that it does not take into account the complex spatial dependencies (inter- and intra-modal correlations) of sensor data. Therefore, we propose a model that can accurately and effectively detect outliers in multi-modal time series collected from manufacturing facilities by utilizing Graph Attention Networks. Our proposed model demonstrates highest performance through performance comparison with six baselines.