Fragility assessment of tunnels in soft soils using artificial neural networks

Abstract Recent earthquakes have shown that tunnels are prone to damage, posing a major threat to safety and having major cascading and socioeconomic impacts. Therefore, reliable models are needed for the seismic fragility assessment of underground structures and the quantitative evaluation of expected losses. This paper builds on previous research and presents a probabilistic framework based on an artificial neural network (ANN), aiming at the development of fragility curves for circular tunnels in soft soils. Initially, a two-dimensional incremental dynamic analysis of the nonlinear soil-tunnel system is performed to estimate the response of the tunnel under ground shaking. The effects of soil-structure-interaction and ground motion characteristics on the seismic response and fragility of tunnels are adequately considered within the proposed framework. An ANN is employed to develop a probabilistic seismic demand model, while its results are compared with the traditional linear regression models. Fragility curves are generated for various damage states accounting for the associated uncertainties. The results indicated that the proposed ANN-based probabilistic framework results to reliable fragility models, having similar capabilities as the traditional approaches, while lower computational cost is required. The proposed fragility models can be adopted for the risk analysis of typical circular tunnel in soft soils subjected to seismic loading, and they are expected to facilitate decision-making and risk management toward more resilient transport infrastructure.