The extrapolation of the whole life load spectrum from the strengthened road load spectrum in the test site is the key factor for the accuracy of the fatigue durability evaluation of automobile components. In this paper, aiming at the limitation of parameter method extrapolation single distribution estimation, the non parametric extrapolation estimation is introduced to describe the comprehensive characteristics of the load spectrum distribution, and the two extrapolation methods are compared parametrically, and the optimal fitting extrapolation spectrum is selected to carry out the fatigue damage analysis of the sub-frame. Based on the real vehicle test in the proving ground strengthened road, the stress spectrum of the dangerous point of the sub-frame is extracted by co-simulation of multi-body dynamics and finite element method. The parametric method and non-parametric method are used to calculate the goodness of fit of stress spectrum, the optimal mean value and amplitude value probability density of stress spectrum are selected according to the goodness of fit, and the stress spectrum cycle was extrapolated to 106 times, and then the stress spectrum was divided into eight levels. According to Miner criterion, Haibach and NASA S-N curve correction models considering small load damage were respectively used to estimate fatigue damage of front sub-frame. The results show that the fitting effect of non-parametric extrapolation is better than that of parametric extrapolation, and the goodness of fit of equivalent amplitude is 98.6%, which improves the description accuracy of stress spectrum distribution characteristics and provides more accurate extrapolation spectrum for fatigue durability evaluation of components. Based on the extrapolated spectrum and using Haibach model and NASA model to study the fatigue damage of sub-frame, it is shown that NASA model is more suitable for reliability design of components, while Haibach model is suitable for lightweight design of components.