(POSTER) Navigating the Unknown: Anomaly Detection in Sensor Nodes Based on Event Traces

In wireless sensor networks, post-deployment issues persist despite extensive testing, primarily due to unpredictable environmental factors and limited debugging tools for resource-constrained end nodes. This challenge is particularly pronounced in remote applications such as extraterrestrial habitats. To address this, we propose a runtime anomaly detection and diagnosis method for resource-constrained sensor nodes. A key advantage of our approach is its ability to learn expected behavior from historical data, eliminating the need for explicit behavior modeling, unlike other runtime fault detection methods. Our method comprises three main components: logging, detection, and diagnosis. We log event traces on the sensor nodes, enabling activity tracking down to the variable level. For anomaly detection, we explore various methods, including state transition, execution interval analysis, and clustering. Subsequently, diagnosis is performed using the logged event traces.