Structural Health Monitoring (SHM) systems help to monitor critical infrastructures (bridges, tunnels, etc.) remotely and
provide up-to-date information about their physical condition. In addition, it helps to predict the structure’s life and
required maintenance in a cost-efficient way. Typically, inspection data gives insight in the structural health. The global
structural behavior, and predominantly the structural loading, is generally measured with vibration and strain sensors.
Acoustic emission sensors are more and more used for measuring global crack activity near critical locations. In this
paper, we present a procedure for local structural health monitoring by applying Anomaly Detection (AD) on strain
sensor data for sensors that are applied in expected crack path. Sensor data is analyzed by automatic anomaly detection
in order to find crack activity at an early stage. This approach targets the monitoring of critical structural locations, such
as welds, near which strain sensors can be applied during construction and/or locations with limited inspection
possibilities during structural operation. We investigate several anomaly detection techniques to detect changes in
statistical properties, indicating structural degradation. The most effective one is a novel polynomial fitting technique,
which tracks slow changes in sensor data. Our approach has been tested on a representative test structure (bridge deck) in
a lab environment, under constant and variable amplitude fatigue loading. In both cases, the evolving cracks at the
monitored locations were successfully detected, autonomously, by our AD monitoring tool.