Anomaly detection in long-term tunnel deformation monitoring

Abstract

Tunnels are critical infrastructure that require a careful monitoring of their structural health. This paper explores the use of novelty detection algorithms for long-term tunnel monitoring projects. The algorithms are applied to the monitoring of the Waasland tunnel in Antwerp, where the deformations and temperatures have been monitored over a period of one year. The case demonstrates that novelty detection by means of principal component analysis enables the identification of minor changes in the tunnel response, and can therefore be embedded in an early detection warning system. An additional numerical analysis based on simulation data further demonstrates that this approach requires a sufficiently long training period, in order to enable distinguishing between regular deviations in the tunnel response and anomalies.