IEEE Access (Jan 2023)
On Anomaly-Aware Structural Health Monitoring at the Extreme Edge
Abstract
Self-awareness has been successfully utilized to create adaptive behaviors in wireless sensor nodes. However, its adoption can be daunting in scenarios, such as structural health monitoring, where the monitored environment is too complex for it to be accurately modeled by a sensor node. This article addresses this challenge by proposing a novel and lightweight anomaly-aware monitoring method for structural health monitoring that can be directly executed by a sensor node. Instead of modeling the complete structure, the proposed anomaly-aware monitoring method uses the vibration measurements of the sensor node to identify local deviations in the dynamic response of the monitored structure. The self-awareness module can then use this information to guide the dynamic behavior of the sensor node, replacing more resource-intensive structural models. We use data from multiple public benchmark structures to evaluate different features and propose an unsupervised feature selection method. Additionally, we evaluate different anomaly detection algorithms comparing their ability to detect local structural damages, also taking into account their memory and energy cost. The proposed method has been implemented in a commercial sensor node, and deployed in a scaled structure where various damage scenarios were simulated to validate the proposed method, where it was able to successfully detect the presence of damages in over 88% of the cases. Finally, we showcase how the proposed method can enhance self-awareness through the use of a simulation, where the proposed monitoring method was able to extend the battery life of the sensor node by over 59%, without impacting the node’s ability to swiftly detect damages in the structure.
Keywords
