Prior Data‐Driven S–N Curves and Fatigue Limit Prediction from Small‐Sample Testing

ABSTRACT To address the challenges of long testing durations, high specimen consumption, and limited prediction capability under small‐sample conditions in high‐cycle fatigue limit evaluation, a small‐sample fatigue limit prediction method based on prior data (P‐BaM) is proposed. The method constructs a mean prior S–N curve using historical tensile and fatigue data from similar materials. A simulated tensile point is introduced to establish the relationship between tensile properties and fatigue performance, and the prior S–N curve is sequentially updated using a limited number of high‐cycle fatigue failure data, enabling fatigue‐limit prediction and test‐strategy optimization. Prior databases were established for 42CrMo alloy steel, LY12 aluminum alloy, 304 stainless steel, and QT400 ductile cast iron, and the proposed method was validated through comparison with the Stromeyer model and a physics‐informed neural network (PINN) model. The results show that the stress at the simulated tensile point exhibits good stability and convergence with respect to tensile properties, providing an effective bridge between tensile and fatigue performance. For the investigated materials, fatigue‐limit prediction of 42CrMo alloy steel, 304 stainless steel, and QT400 ductile cast iron required only two to three fatigue specimens, while six to eight specimens were required for LY12 aluminum alloy to satisfy the stopping criterion and complete the prediction. The specimen requirement is substantially lower than that of conventional fatigue‐testing approaches. The proposed method enables rapid fatigue‐limit prediction using limited fatigue data and provides an effective approach for small‐sample fatigue assessment of similar metallic materials.