Research progresses of prediction method and uncertainty of train-induced environmental vibration

Various prediction methods and their uncertainty problems for the train-induced environmental vibration were summarized systematically. Various methods and models developed in the recent decade were summarized in three prediction classes such as the scoping prediction, determined prediction and detail prediction. The stochastic uncertainty of model input parameters was discussed, including the uncertainties caused by the difference between trains, the wheel and rail wear growth, and the input soil parameters in the prediction model. Based on the new measurement results, the influence of wheel and rail wear state on the uncertainty of vibration source intensity of metro was analyzed. Research result indicates that the machine learning method and analytical transfer function method of soil layers can be employed in the scoping prediction stage. In the determined prediction stage, various numerical and analytical models improve gradually, and their prediction efficiencies increase gradually. However, the wheel-rail excitation input method considering the developments of wheel and rail wear still needs to be further improved. The dynamics model of building structure with clear vibration propagation path and for the engineering prediction should still be further developed. In the detail prediction, the hybrid prediction method needs to be developed, and the application on the vibration prediction for underground metro lines should be investigated. Up to now, there is lack of researches on the accuracy and precision of prediction result and the reliability of prediction method. Most predictions can only provide results with determined values. The uncertainties of wheel and rail wear, operation maintenance level and vibration propagation in the soil layers cannot be considered. It is suggested to further develop the micro vibration acquisition device with remote intelligent offline sampling function, and it can be permanently installed on the building structures, so as to combine this technique with the machine learning method to adapt to the development requirements of intelligent prediction in future. It is suggested to develop the roughness spectrum that can describe the rail short-wave wear state grade and wheel out-of-round grade, and establish a dynamic environmental vibration prediction model in a whole maintenance and repair cycle. The intelligent prediction method with reliability and accuracy requirements should be developed, and the prediction should be fundamentally changed from the fixed value prediction to the probability prediction in future.