Probabilistic Risk Assessment of Tunnel Seismic Damage Under Physically Based Non-Stationary Earthquakes

The seismic performance of tunnel structures is significantly influenced by the randomness of ground motions. Traditional probabilistic risk assessments, which rely on limited recorded ground-motion data, often suffer from small-sample bias and fail to capture the full distribution of seismic input. To overcome this limitation, this study employs a physically based stochastic ground-motion model to generate a large and statistically representative sample ensemble. A probabilistic seismic risk assessment framework is then developed using the stochastic finite element method, explicitly incorporating ground-motion uncertainty. Four statistical criteria, namely practicality, correlation, efficiency, and proficiency, are systematically applied to evaluate candidate intensity measures (IMs) and identify the optimal one. Among all candidates, PGA exhibits the best overall performance, with the highest regression fitness (R² = 0.873) and the lowest dispersion (β_D = 0.197), followed by PGV (R² = 0.848, β_D = 0.215). Fragility curves for different damage states are subsequently derived. Results indicate that structural responses vary considerably under stochastic ground-motion excitation, and the failure probability follows a typical S-shaped curve as intensity increases. Moreover, the failure probabilities for different damage states exhibit nonlinear growth at higher intensity levels. These findings provide a mathematical basis for probability-based seismic design and risk assessment of tunnel structures.