Information theoretical noninvasive damage detection in bridge structures

TL;DR

This paper explores noninvasive damage detection in bridges using information-theoretic analysis of sensor data, revealing changes in mutual information and interaction patterns indicative of structural damage.

Contribution

It introduces a novel application of information-theoretic measures and a pruning method to detect and analyze damage in bridge structures from sensor data.

Findings

Sensor acceleration data follow Laplace distribution, enabling specific estimators.

Mutual information between sensors decreases with damage, indicating loosened structure.

Interaction directions align with traffic flow, even after damage.

Abstract

Damage detection of mechanical structures such as bridges is an important research problem in civil engineering. Using spatially distributed sensor time series data collected from a recent experiment on a local bridge in upper state New York, we study noninvasive damage detection using information-theoretical methods. Several findings are in order. First, the time series data, which represent accelerations measured at the sensors, more closely follow Laplace distribution than normal distribution, allowing us to develop parameter estimators for various information-theoretic measures such as entropy and mutual information. Secondly, as damage is introduced by the removal of bolts of the first diaphragm connection, the interaction between spatially nearby sensors as measured by mutual information become weaker, suggesting that the bridge is "loosened". Finally, using a proposed oMII procedure to prune away indirect interactions, we found that the primary direction of interaction or influence aligns with the traffic direction on the bridge even after damaging the bridge.