The test data of smooth specimens is expanded, by integrating Bayesian theory and the Monte Carlo method, and a probabilistic fatigue model considering size effect under different loading conditions is established. Furthermore, the influence of non-uniform stress fields and notch characteristics on fatigue crack initiation is defined, considering critical distance, with a correction to the stress field damage parameter. The relationship between fatigue life and damage variables under various loading conditions is explored, the virtual sub-sample augmentation method and the weakest link theory are combined, leading to the establishment of a multiaxial fatigue life prediction model, suitable for notched specimens. Finally, the prediction life using proposed method and other models are compared with the experimental life of three materials, the compared results show that the proposed method exhibits a notably heightened level of accuracy, and the fatigue life of notched specimens can be predicted by small samples of smooth specimens.