Obtaining the dynamic stress spectrum is an important prerequisite for vehicle structural fatigue analysis. The accuracy of analysis results is greatly affected by the integrity and sufficiency of the dynamic stress spectrum. Therefore， it is necessary to extrapolate the dynamic stress test data from a small sample to a large sample under the premise of preserving the probability distribution characteristics of the load. In this paper， the stress spectrum extrapolation and fatigue life assessment of the bogie frame are studied， and a test point is taken as an example. First， the determination method of optimal bandwidth in two-dimensional kernel density estimation is discussed. Second， the matrix grey-correlation analysis method is proposed for the first time to evaluate the proximity-correlation degree and similarity-correlation degree of the rain-flow matrix before and after extrapolation. Finally， a fatigue life assessment method based on a multi-sample dynamic stress spectrum kernel density extrapolation is proposed. The results show： compared with the actual data， the error of fatigue life assessment based on kernel density extrapolation is -4.17%， that is， the safe operation mileage is reduced by 21?300?km. It is proved that the extrapolation based on kernel density estimation can meet the accuracy requirements of engineering， and the evaluation results are safer， which is very beneficial to ensure the safe operation of vehicles.