Obtaining a representative dynamic stress time-domain signals is a prerequisite for structural fatigue analysis of vehicles. However， the fatigue life assessment results only depend on the dynamic stress time-domain signal obtained by a short-time test； it can not represent the anti-fatigue service performance of the structure in a long-time， even in the whole service life. Therefore， a fatigue life assessment method based on time-domain extrapolation of dynamic stress based on extreme theory is proposed for the first time. Taking a test point near the weld of the bogie arm positioning mounting seat of an electric multiple units （EMU） as an example， first， three groups of dynamic stress time-domain signals of the test point at the initial stage， middle stage， and the final stage of a whole wheel reprofile period are selected. The pre-processing and equal weight overlapping addition are carried out to them. Second， the tail probability distribution of the superimposed signal samples is fitted using the extreme value theory. The fitting goodness of the model is tested. Third， according to the fitting probability distribution function， the dynamic stress samples are extrapolated in the time domain to obtain the dynamic stress time-domain signal under long-term load； finally， the fatigue life of the test point is evaluated according to the extrapolated dynamic stress time-domain signal. The results show that： compared with the commonly used linear extrapolation method， the cumulative fatigue damage based on the time-domain extrapolation method increases by 0.168%， the safe operation mileage is reduced by 2 670 km， and the assessment is more safety.