Aiming at the problem of damage accumulation caused by the impact of enemy weapons and their own artillery recoil during naval battles, a method of impact fatigue life prediction combining membership functions and Miner damage accumulation theory is proposed. The impact amplitudes and means are extracted from external shocks, and the parameters of shock amplitudes are predicted according to GM(1, 1) model, and the residual errors are corrected by BP neural network to improve the prediction accuracy. A new stress-life model is proposed based on the relationship between fatigue modulus and life, and the parameters in the model are determined based on polynomial variation theory. Taking 6061 aluminum alloy shafts, the key parts of warship equipment, as the test object, and combining with several membership functions, the impact fatigue lives of the specimens are predicted. The results show that the errors of the predicted values of impact fatigue lives obtained by the membership function of Γ distribution are the smallest compared with the experimental values, and when the value of k in Γ distribution is constant, the change of parameter a1 almost has no influence on the predicted value in the initial stage, and only has a slight influence when a1 approaches S0 gradually. It shows that the proposed fatigue life prediction method is reliable, and provides theoretical reference and data basis for the impact damage accumulation research and life prediction of key parts of warship equipment.