Reliability assessment of vibration damping track system based on integrated improved SE-U-Net and PDEM

In the design and management of vibration-damping track systems, it becomes crucial to assess the reliability of vibration-damping effects under the influence of various random variables. To address the stochastic nature of environmental vibration excitation, this study proposes a reliability assessment framework for the PDEM (Probabilistic Dynamic Equilibrium Model) based on a deep learning surrogate model. The surrogate model is constructed by combining the Self-Enhancing Strategy (SE) and the U-Net neural network model, with a loss function that integrates MSE (Mean Squared Error) and PCCs (Pearson Correlation Coefficients). This enhances the surrogate model's ability to dynamically extract channel features, enabling rapid predictive analysis of the vibration-damping effects of railway traffic source strengths. Then combines multiple prediction results from the surrogate model with time series data and Latin Hypercube Sampling to quickly assess the reliability of vibration-damping effects in railway traffic. Finally, a case study of the vibration-damping renovation of a metro trapezoidal tie is presented. Through comparison with on-site measured data and the classical Monte Carlo method, the effectiveness and computational efficiency of the surrogate model are demonstrated. Compared to traditional PDEM methods based on vehicle-track coupled dynamics mode, the proposed approach improves computational efficiency by a factor of 1130 to 1500. This research provides an adaptive vibration prediction tool for intelligent rail systems and a multi-condition rapid reliability optimization platform.