Novel Context-Aware Multivariate Time Series Modelling for Underground Transportation Infrastructure Monitoring and Management

The project aims to develop and implement novel AI techniques on smart sensing systems to address an unmet need in the transport sector for a scalable and effective means to detect, diagnose and predict potential faults in high-risk underground transportation infrastructure.

Target Sector: Transportation

Lead PI: Prof Yang Yaowen (NTU)

Co-PIs

  • Prof Cong Gao (NTU)
  • Assistant Prof Fu Yuguang (NTU)

Collaborator

Kensuke Date (KAJIMA Technical Research Institute Singapore)

Host Institution: Nanyang Technological University (NTU)

Industry Partners:

  • Land Transport Authority (LTA)
  • KAJIMA Technical Research Institute Singapore

Problem Scope

Tunnels constitute an important component of Singapore’s railway network, of which a significant proportion of ~45% is located below the ground. However, condition assessment and maintenance of these tunnels are challenging due to the complex and harsh underground environment, continuous high-rate dynamic loads, and limited access time for inspection.

Design Approach

The project leverages a network of distributed fibre optic sensors as main data collection instruments and other sensors e.g. wireless accelerometers which measure strain, temperature, vibration, acceleration etc. autonomously and continuously. To utilise these context-aware multivariate time series data generated by the sensors, we developed two key AI techniques: (1) effective data imputation techniques for context-aware time series data reconstruction from the measurements with various anomalies e.g. missing, noisy, drift etc. and (2) built upon the reconstructed data, effective and explainable deep learning models for context-aware fault diagnosis and prediction in tunnel networks.

Potential Impact/Benefits to Target Sector

The proposed AI-based smart sensing system will address the needs of tunnel owners and engineering services companies by providing them with actionable information regarding accurate and reliable detection and prognosis of potential faults in large-scale tunnel networks. The developed framework will enable (i) real-time monitoring and condition assessment of tunnels under both sudden events e.g. ground tremors and long-term deterioration e.g. ground deformation, and (ii) predictive maintenance of subsurface infrastructure.

The smart infrastructure management system presented in this project will reduce disruptions and downtime caused by sudden tunnel failures leading to economic loss. Together, these will enhance the overall safety, efficiency, and disaster resilience of Singapore’s underground tunnel network leading to prolonged service life.