Monitoring of Bridges by MT-InSAR and Unsupervised Machine Learning Clustering Techniques

Continuous monitoring of critical infrastructures is crucial to prevent catastrophic events such as collapse of viaducts and prioritising maintenance interventions. However, developing effective monitoring approaches must rely on the collection of numerous information, such as the time series of structural deformations. In this context, various ground-based non-destructive testing (NDT) methods have been used in monitoring the structural integrity of transport infrastructures. These require routine and systematic application at the network level over long periods of time to build up a solid database of information, involving many efforts from stakeholders and asset owners in the sector. To this effect, satellite based remote sensing techniques, such as the Multi-temporal Interferometric Synthetic Aperture Radar (MT-InSAR), have gained momentum due to the provision of accurate cumulative structural displacements in bridges. Although the application of the InSAR monitoring technique is established, this is limited by the high amount of time required for the interpretation of data with high spatial and temporal density. This research aims to demonstrate the viability of MT-InSAR techniques for the structural assessment of bridges and the monitoring of damage by structural subsidence, using high-resolution SAR datasets, integrated with complementary Ground-Based (GB) information. To this purpose, high-resolution SAR dataset of the COSMO-SkyMed (CSK) mission provided by the Italian Space Agency (ASI), were acquired and processed in the framework of the ASI-Open Call approved Project "MoTiB" (ID 742). To elaborate, a Persistent Scatterer Interferometry (PSI) analysis is applied to identify and monitor the structural displacements at the Rochester Bridge, in Rochester, Kent, UK. To explore the viability of Machine Learning algorithms in detecting critical scenarios in the monitoring phases, an Unsupervised ML Clustering approach, which generates homogeneous and well-separated clusters, is implemented. Each PS data-point is allocated to specific cluster groups, based on individual deformation trend features and the values of displacements from the historical time-series. This research paves the way for the development of a novel interpretation approach relying on the integration between remote-sensing technologies and on-site surveys to improve upon current maintenance strategies for bridges and transport assets.